Chapter 5. Measuring Nonpollution Externalities from Motor Vehicles
  • 1 0000000404811396 Monetary Fund
  • | 2 0000000404811396 Monetary Fund
  • | 3 0000000404811396 Monetary Fund
  • | 4 0000000404811396 Monetary Fund


This chapter consists of three sections focused on the three major, non-pollution-related externalities from motor vehicles: traffic congestion, traffic accidents, and (to a much lesser extent) wear and tear on the road network (relevant for trucks). Other data and assumptions needed to implement the corrective motor fuel tax formulas from Chapter 3 are discussed in the annexes to this chapter.

This chapter consists of three sections focused on the three major, non-pollution-related externalities from motor vehicles: traffic congestion, traffic accidents, and (to a much lesser extent) wear and tear on the road network (relevant for trucks). Other data and assumptions needed to implement the corrective motor fuel tax formulas from Chapter 3 are discussed in the annexes to this chapter.

Congestion Costs

Basically, what is needed here is the cost of reduced travel speeds for other road users caused by an extra kilometer of driving by one vehicle, averaged across different roads in a country and across times of day. This cost estimate can then be used in the formula for corrective motor fuel taxes (equation (3.1)). As noted in Chapter 3, to manage congestion on the road network most effectively, countries should ideally transition to kilometer-based taxes that vary with the prevailing degree of congestion on different roads. However, until these schemes are comprehensively implemented, charging motorists for congestion costs through fuel taxes is entirely appropriate.

The congestion cost has two main components: First is the average added travel delay to other road users, as defined more technically in Annex 5.1, which (because of lack of direct data) needs to be extrapolated to the country level. Second, to convert delays into a monetary cost is the value of travel time (VOT), which is related to local wage rates.

The chapter begins by using a city-level database (covering numerous countries) to establish statistical relationships between congestion delays and various transportation indicators. These results, and country-level data for those same indicators, are then used to extrapolate congestion delays to the country level. Next, the way in which delays are converted into congestion costs is discussed. Results are presented, then a quick check on the results is performed by comparing them with cost estimates obtained from detailed country-level data (for a couple of countries for which these data are readily available).

The focus is on the most important cost component (time lost to motorists). Box 5.1 reviews some broader costs that should, in principle, be factored into corrective fuel tax assessments, but that are beyond the scope of this volume. For this reason, along with other assumptions made below, the congestion cost estimates in this chapter are probably on the low side.

Broader Costs of Congestion

One additional cost, beyond the pure time losses from travel delay, is the added fuel cost to other motorists from the possible deterioration in fuel efficiency experienced under congested conditions. The link between slower travel speeds and fuel consumption rates is complicated, however (Greenwood and Bennett, 1996; Small and Gómez-Ibáñez, 1998). Sometimes congestion slows traffic without increasing stop-and-go conditions, which could improve fuel efficiency for some range of relatively fast travel speeds. For the United States as a whole, Schrank, Lomax, and Eisele (2011, p. 5) estimate added fuel costs to be about 5 percent of the total costs of congestion, suggesting that these costs may have only modest implications for corrective fuel taxes.

Other, more subtle, costs of congestion may be more significant. For example, people may choose to set off earlier or later to avoid the peak of the rush hour, which may cause them to arrive earlier or later at their destination than they would otherwise prefer, perhaps because early arrival means they waste time waiting for an appointment, or late arrival runs the risk of penalties at work. Furthermore, congestion can result in day-to-day uncertainty about travel times, making it more difficult to plan the day (e.g., scheduling appointments, dinner times, and day care pickups). Studies suggest that travel time variability alone might raise the overall costs of congestion by about 10–30 percent (Eliasson, 2006; Fosgerau and others, 2008; Peer, Koopmans, and Verhoef, 2012).

Travel Delays at the City Level

The starting point is the Millennium Cities Database for Sustainable Transport, which provides detailed information on transportation in 100 cities (10 are discarded because of missing data).1 These cities are listed in Annex 5.2.

The data are for 1995 and therefore rather dated, though they are the best available. Moreover, the age of the data need not be a problem because they are used for estimating statistical relationships between travel delay and transportation indictors, which are then matched with recent, country-level data on those indicators to provide up-to-date country-level estimates of travel delay. This approach is reasonable so long as the statistical relationships between delays and transportation indicators have not changed substantially since 1995.

The average road network speed in the database is the average speed of all motor vehicles (average for 24 hours/day, 7 days/week) on all classes of road in the metropolitan area.2 These data provide information on recurrent congestion delays (occurring each day under normal driving conditions) but not on the average amount of nonrecurrent congestion (occurring from sporadic events such as accidents, bad weather, and road work). In this sense congestion costs are understated, perhaps significantly.3

As indicated in Table 5.1, across all cities the average travel speed is 34.2 kilometers/hour, with speeds well above this average in North American cities (47.7 kilometers/hour) and well below it in non-affluent Asian cities such as Delhi (20.6 kilometers/hour).

Table 5.1.

City-Level Travel Delays and Other Characteristics, Region Average, 1995

article image
Sources: Millennium Cities Database; and for average delay, authors’ calculations.Note: Figures are simple averages across urban centers in different regions. The average road network speed is the average speed of all vehicles (24 hours/day; 7 days/week) on all classes of road in the metropolitan area.

The speed data are used to derive average travel delays using assumptions about travel speeds that would occur in the absence of congestion.4 As indicated in Table 5.1, these estimated average delays per vehicle-kilometer are lowest in North America (0.006 hours/kilometer), and more than twice as large in western and eastern Europe, the Middle East, Africa, and affluent Asian cities such as Tokyo. Delays per kilometer are greater still in Latin American cities and non-affluent Asian cities.

Statistical regressions are used to obtain a relationship for predicting average delays for countries as a whole, using common indicators that are available both for the 90 cities in the Millennium Cities Database and in the country-level data discussed below. These variables comprise the following:

  • Metropolitan GDP per capita (an indicator of a city’s level of economic development)

  • Annual car-kilometers (an indicator of traffic mobility)

  • Road length or capacity per car

  • Cars in use per capita (this, and the previous variable, are indicators of traffic intensity, relating to transport infrastructure and supply).

Common statistical techniques are used to estimate coefficients that show the contribution of each of these indicators toward explaining average travel delays across cities, using functional forms that best fit the data. Further details, along with the statistical regression results, are provided in Annex 5.3.

Ideally, additional variables would be included in these regressions to improve statistical accuracy. However, because the purpose is to make country-level extrapolations, only those indicators for which data are available at the country level can be used. Despite this limitation, a reasonably good statistical fit is still obtained.

Projecting Country-Level Delays

The estimated statistical relationships between the average delay and the four key indictors at the city level are now used to project the average delay for 150 countries (all countries for which these data are available), with the country-level indicators. For this purpose, GDP per capita is taken from World Bank (2013) and all other indicators from World Road Statistics 2009 (IRF, 2009).5 For 81 of the countries, data on car-kilometers traveled are missing. Annex 5.3 describes how this data gap was filled using supplementary statistical regressions.

Table 5.2 summarizes the key indicators by region. At the country level, per capita incomes are lower, annual kilometers driven per car are higher, and road capacity per car is smaller than in the city-level data shown in Table 5.1.

Table 5.2

Country-Level Travel Delays and Other Characteristics, Region Average, 2007

article image
Sources: IRF (2009); and authors’ estimates of some data on annual km driven per car (see Annex 5.3). Average delay is predicted using procedures described in the text.Note: km = kilometer. Amounts are simple averages across countries—therefore, for example, high average delays in Mexico inflate the average amount for North American countries.

The estimated coefficients from the city-level analysis are used together with country-level variables to predict the average nationwide delays in the 150 countries. Because the city-level regression is based on 90 major cities, the average predicted delay represents the urban congestion level for each country, excluding the rural areas. To predict the average delay at the country level, the predicted urban average delay is scaled by the urban population ratio, on the assumption that rural congestion is negligible.

Comparing the results in the second column of Table 5.2 with those from Table 5.1, the average vehicle delays at the country level are about one-quarter to one-half of those at the city level. This difference makes sense—the city level data focus only on delays in large cities (where congestion is especially severe), whereas the country-level estimates also account for driving in rural areas and medium and small cities. Nonetheless, it is important to bear in mind that average travel delays at the country level are estimated with a fair amount of imprecision, especially for countries for which there might be substantial errors in the measurement of transportation indicators.6

From Delays to Congestion Costs

This section explains how delays that one vehicle imposes on others are derived from the above estimates and then monetized. Complications posed by other vehicles on the road, such as buses, are also discussed.

Deriving delays imposed on others by one vehicle

A specification commonly used by transportation engineers for the relationship between travel speed or time and traffic volume results in a simple relationship between average delays per kilometer (estimated above) experienced by individual drivers and the increased travel time that one extra vehicle implies for all other vehicles on the road.

When travel delay is a simple power function of traffic volume relative to road capacity, with the exponent in this function denoted by β, then the extra delay one vehicle imposes on other vehicles is simply β times the average delay per kilometer (see Annex 5.1). Empirical studies suggest that β is roughly in the range of 2.5–5.0, with higher values in this range applicable to larger urban centers. In this analysis, β is assumed to be 4.7

Finally, delays to other passengers are obtained by multiplying delays to other vehicles by the vehicle occupancy rate, assumed to be 1.6 (Annex 5.4).8 Alternative assumptions about vehicle occupancy and the exponent β would have proportional effects on the congestion costs reported below (e.g., if β = 5 or average vehicle occupancy is 2, congestion costs would be 25 percent greater).

Value of travel time (VOT)

The discussion now turns to the VOT, which is needed to monetize congestion costs.

According to economic theory (Becker, 1965), on average, people should organize their time such that they are indifferent between an extra hour at work and an extra hour of nonmarket time (e.g., relaxing at home, looking after the children). Therefore, an extra hour of nonmarket time is commonly valued by the benefit to individuals of an extra hour of forgone work, namely, the after-tax hourly wage (i.e., the market wage after netting out personal income and employee payroll taxes, and consumption taxes paid when wages are spent).

As a first pass, people might also value an extra hour of travel time by the net-of-tax wage, which would suggest a VOT of about 50–70 percent of the market wage for a typical advanced country. More generally, the monetary cost of travel time could be lower (if people enjoy driving, for example, because they can listen to music) or higher (if people enjoy the workplace, for example, because of interaction with colleagues).

A large empirical literature estimates the VOT for personal travel using revealed and stated preference techniques similar to those discussed in Chapter 4. A revealed preference study might involve, for example, estimating people’s willingness to pay extra auto fuel and parking costs to save time as compared with an alternative, slower travel mode, while a stated preference study might involve directly asking people what tolls they might pay for a faster commute.

For Canada, France, the United Kingdom, and the United States, literature reviews suggest that a VOT of about half the market wage is a reasonable rule of thumb for general automobile travel (see Table 5.3). The VOT is somewhat higher for commuting (e.g., because of penalties for late arrival at work) than for non-market-related trips such as shopping, taking the children to school, or going to the gym—16 percent higher according to Wardman (2001). Here the VOT is assumed to be 60 percent of the market wage, given that most delays occur during the commuter-dominated peak period.

Table 5.3

Reviews of Empirical Literature on the Value of Travel Time (VOT)

article image
Note: Summary findings for these reviews were taken from Small and Verhoef (2007, pp. 52–53). Studies take a weighted average over different trip types (usually at peak period) except for Waters (1996) who focuses exclusively on commuter trips.

The VOT-to-market wage ratio is assumed to be the same across all countries.9 The wage data are from the International Labor Organization’s Global Wage Database (ILO, 2012) and are nationwide measures for 2010.10

Figure 5.1 shows the VOT for selected countries. Broadly speaking, the relative pattern of VOTs across countries is similar to that for the value of mortality risks in Figure 4.2 of Chapter 4.11

Figure 5.1
Figure 5.1

Value of Travel Time, Selected Countries, 2010

Source: Authors’ calculations.

Accounting for other vehicles

The estimates in this analysis assume that all vehicles on the road are cars, whereas in practice the vehicle fleet comprises a mixture of cars, buses, trucks, and two-wheel motorized vehicles. Annex 5.4 discusses and applies a formula that shows the ratio of congestion costs (properly estimated accounting for the mix of vehicles) relative to the congestion cost estimated here.

If trucks and two-wheelers account for a sizable portion of the vehicle fleet (but buses do not) the estimates are not very different. However, if buses account for a significant portion of vehicle-kilometers, the estimates here can substantially understate congestion costs (see the Annex 5.4). Cars have a significantly greater impact on increasing travel times for other road users when a greater portion of vehicles on the road are carrying large numbers of passengers. An adjustment is not made here, however, because data on the share of buses in urban vehicle-kilometers are not available for many countries.12

Finally, in the computation of corrective diesel fuel taxes, an extra truck-kilometer is assumed, based on the literature (Lindsey, 2010, p. 363; Transportation Research Board, 2010; Parry and Small, 2009), to contribute twice as much to congestion as an extra car-kilometer (trucks drive more slowly and take up more road space, though a partially offsetting factor is that they tend to be driven less intensively on congested roads).


Figure 5.2 shows nationwide congestion costs imposed on others per extra car-kilometer, for 20 selected countries.

Figure 5.2
Figure 5.2

Congestion Costs Imposed on Others per Car-Kilometer, Selected Countries, 2010

Source: Authors’ calculations.

The congestion cost for the United Kingdom, for example, is US$0.09/kilometer. Australia, Germany, Israel, Korea, and South Africa all have broadly similar congestion costs, while Turkey’s is substantially higher, and Japan’s higher still. (Although Japan has a relatively high VOT, most of the difference is due to its greater estimated travel delays.) Congestion costs for the United States, where a smaller portion of nationwide driving occurs under congested conditions, are lower at US$0.064/ kilometer (though this U.S. estimate seems on the high side relative to a potentially more accurate estimate discussed below). China’s estimated congestion cost is US$0.05/kilometer, less than the United States—despite China having greater average travel delays—because of its much lower assumed VOT. Low VOTs also help explain the low congestion costs (less than US$0.01/kilometer) in India and Kazakhstan.

Figure 5.3 shows ranges of estimated congestion costs for all countries, where data allow. Again, these costs are relatively high in western Europe (where a large portion of driving occurs under congested conditions and people have a high VOT) and, except for South Africa, relatively low in Africa (where the VOT is lowest). The United States, Latin America, and Australia are intermediate cases.

Figure 5.3
Figure 5.3

Congestion Costs Imposed on Others per Car-Kilometer, All Countries, 2010

Source: Authors’ calculations.

Robustness Checks

For the United Kingdom and the United States, detailed data on travel delays for road classes in different regions are available and can be combined into an alternative estimate of nationwide average delay as a check on the above estimates (see Annex 5.5 for estimation procedures and data sources).

For the United Kingdom, the average delay per vehicle-kilometer obtained from this alternative data source is almost exactly (within 1 percent of) that estimated above, providing some reassurance that the approach, at least for the United Kingdom, might be reasonable. For the United States, the average delay using the alternative data is 59 percent of that estimated above, suggesting that the estimate in this analysis may be on the high side for that particular country.

The delay estimates from country-level data should be more reliable than the extrapolations presented above, though the former are surprisingly hard to come by (transportation authorities do not routinely collect these data). The approach used in this chapter suffers from imprecision given the limited number of indicators common to both the city- and country-level data, issues with the quality of both city- and country-level data sets, and the possibility that the underlying relationship between the average delay and city or country characteristics might have evolved with changes in infrastructure, technology, and traffic rules. However, it is hard to gauge the direction, let alone the magnitude, of bias for individual countries. Moreover, broadly speaking the pattern of relative congestion costs across different countries estimated above seems plausible, even though the individual country estimates may not be especially accurate.

Accident Costs

The total societal costs from road traffic accidents can be substantial, and are often underappreciated. Gauging the appropriate charge for accident risk to be reflected in fuel taxes as a complement to other measures such as road safety investments is difficult, however, for two reasons.

First, conceptually it is a bit tricky to judge whether certain categories of costs should be viewed as “internal” because individuals take them into account in their driving decisions, or “external,” that is, borne by others. (Only the latter warrant corrective taxes.)

Second, although data are often available for road fatalities, they are usually not available for other accident costs such as nonfatal injuries, medical and property damage costs, and even fatality data are not always broken down in a way that permits assessment of external costs.

The estimates in this chapter necessarily rely on some judgment calls, extrapolations to fill in missing costs, and transfers of fatality breakdowns across similar countries. For these and other reasons, the accuracy of cost estimates can be questioned. But again, the estimates provide some plausible and transparent sense of external accident costs, shed light on why these costs differ across countries, and highlight the data needed to improve the future accuracy of cost assessments.

The discussion proceeds as follows: conceptual issues are reviewed in an attempt to categorize different accident costs into internal versus external risks; the estimation of external costs is discussed; then results are presented.

Classifying Accident Risks: Some General Principles

The main societal costs of road accidents include personal costs of fatal and nonfatal injuries, medical costs, and property damage.13


Injury risks to pedestrians and cyclists, to vehicle occupants in accidents involving only a single vehicle, and to vehicle occupants in accidents involving multiple vehicles are considered separately.

Pedestrian and cyclist injuries: It is normally assumed that motorists do not take into account injury risks they pose to pedestrians and cyclists when deciding how much to drive (Newbery, 1990; Parry, 2004).14 Such risks are therefore classified as external.

Injury risk to occupants in single-vehicle collisions: For accidents involving one vehicle, it is standard to view the injuries to occupants of such vehicles as risks that are taken into account: in other words, if individuals put themselves at greater risk (by getting in the car more often), this is not viewed as a basis for taxation to deter this behavior.15 For similar reasons, injury risks to other occupants (e.g., family members) in single-vehicle collisions are generally viewed as internalized risks.

Injury risks to occupants in multivehicle collisions: Here the delineation between internal and external risks becomes murky. The issue is how extra driving by one vehicle affects injury risks to occupants of other vehicles. All else the same, extra driving by one motorist leads to more cars on the road and greater risks to others—cars have less road space on average and are therefore more likely to collide. In this case, injury risks to other vehicle occupants would increase approximately in proportion to the amount of traffic.

However, all else might not be the same: with more vehicles on the road, motorists may drive more carefully or be obliged to drive more slowly. Thus, an offsetting reduction in accident frequency and in the average severity of injuries in a given accident (because vehicles collide at slower speeds) might occur. Although slightly slower driving may not do much to reduce injury risks to unprotected pedestrians, the effect may be more pronounced for other vehicle occupants, who have a greater degree of protection. What matters, then, is the impact of additional driving on the “severity-adjusted” injury risk to other vehicle occupants. However, available evidence is inconclusive.16

An intermediate assumption between the two more extreme cases is considered in this analysis. In one case, additional driving leads to a proportionate increase in injury risks to others (there is no offsetting decline in severity-adjusted accident risk due to slower or more careful driving). In the second case, extra driving has no effect on severity-adjusted injury risks to others; increased risk to others is completely offset by a decline in the average severity of injuries.

In the first case, it is assumed that half of injuries in multivehicle collisions are external based approximately on the logic that, on average, one vehicle is responsible for the collision and another is not and that those at fault take into account risks to occupants of their vehicles but not occupants of other vehicles (Parry, 2004). In the second case, all injuries in multivehicle collisions are internal. Splitting the difference suggests that one-quarter of multivehicle collision injuries should be treated as external.

Medical and property damage costs

Medical costs associated with all traffic-related injuries are largely borne by third parties (the government or insurance companies), though individuals typically bear some minor portion of these costs through, for example, copayments and deductibles.

It is difficult to pin down how much property damage, primarily repairs or replacement costs for damaged vehicles, drivers take into account. In countries with comprehensive insurance systems, some costs are borne by third parties (insurance companies) but other costs are borne by drivers in the form of deductibles and possibly elevated future premiums following a crash.17

Accident risks from heavy vehicles

Accident costs for trucks are needed to compute the corrective diesel fuel tax. At first glance, it might appear that trucks would impose much greater risks to other road users than cars, given their much greater weight. A counteracting factor, however, is that trucks are driven at slower speeds than cars and that truck drivers are professionals, which may further reduce their crash risk (e.g., because truck drivers are unlikely to drink and drive).18 According to a detailed study by the United States Federal Highway Administration (US FHWA, 1997, Table V-24), overall external accident costs per vehicle-kilometer are only slightly higher for heavy vehicles than for cars—therefore, these costs are assumed in this analysis to be the same for cars and trucks (see also Parry and Small, 2009).

External Cost Assessment

IRF (2012) provides data on traffic fatalities for 2010 or the latest available for most countries, based on local data, such as from police reports. WHO (2013) provides data on the breakdown of fatalities by pedestrians, cyclists, occupants of motorized two- to three-wheelers, occupants of four-wheelers, and a miscellaneous category (e.g., bus riders). In cases in which only total fatalities are reported, the breakdown is assumed to be the same as in another, similar country in the same region. The data used here likely underreport, perhaps substantially, road fatalities for many developing countries, providing yet another reason the corrective fuel tax estimates presented later might be understated.19

The vehicle occupant data do not separate out deaths in multivehicle collisions from those in single-vehicle collisions. Based on a simple average across five country case studies (discussed in Annex 5.6), 57 percent of fatalities of occupants of two-, three-, and four-wheelers are assumed to occur in multivehicle collisions. And from the previous discussion, 25 percent of these are external fatality risks, as are all of the pedestrian and cyclist fatalities. The same values by country as used in Chapter 4 for pollution deaths are used to monetize these fatalities.20

Data are not available for most countries for other components of external accident costs—nonfatal injuries, medical costs, and property damage. However, based on comprehensive estimates of these costs for Chile, Finland, Sweden, the United Kingdom, and the United States, a relationship between the ratio of these other external costs21 to the external costs of fatalities was estimated as a function of the share of external fatalities in total fatalities (Annex 5.6); in countries with a high incidence of pedestrian deaths, the relative size of other external costs tends to be smaller. The external cost ratio was then inferred for different countries based on their shares of external fatalities in total fatalities, and the external costs were scaled up accordingly.


Figures 5.4 and 5.5 show, respectively, the external accident costs for selected countries and for all countries expressed, to facilitate comparison with congestion costs per vehicle-kilometer of travel by car or truck. (See Annex 5.3 on measurement of vehicle-kilometers.)

Figure 5.4
Figure 5.4

External Accident Costs per Vehicle-Kilometer, Selected Countries, 2010

Source: Authors’ calculations.Note: Figure shows external accident costs (reflecting fatal and nonfatal injuries, medical costs, and property damage) expressed per kilometer driven by cars or trucks.
Figure 5.5
Figure 5.5

External Accident Costs per Kilometer Driven, All Countries, 2010

Source: Authors’ calculations.

Higher-income countries tend to have a lower incidence of injuries per kilometer driven because as countries develop, vehicle and road safety tend to improve, and the ratio of pedestrians and cyclists to motorists declines (Kopits and Cropper, 2008).22 This lower incidence of injuries is partially, but not entirely, offset by higher valuations of fatality and injury risk in higher-income countries. Loosely speaking, therefore, these figures show a pattern, with some exceptions, of lower external accident costs per kilometer in higher-income countries. For example, costs are significantly less than US$0.04/kilometer in Australia, Japan, western European countries, and the United States, and more than US$0.06/kilometer in some Central and South American and African countries, and in India, Kazakhstan, and Russia.

Note also (comparing Figures 5.2 and 5.3) that external accident costs can be of the same broad order of magnitude as congestion costs. In fact, in 11 of the selected countries, accident costs are greater than congestion costs.

Road Damage Costs

Vehicle use causes an additional adverse side effect through wear and tear on the road network. However, given that road damage is a rapidly rising function of a vehicle’s axle weight, nearly all of the damage is attributable to heavy-duty vehicles; road damage costs for light-duty vehicles make little difference for corrective fuel taxes (US FHWA, 1997) and are ignored in this analysis.23 Road damage costs also vary considerably across different classes of trucks, which would matter for the design of a finely tuned system of axle weight tolls. However, the concern in this analysis is with damage caused by trucks as a group, so that it can be factored into the corrective diesel fuel tax.

Road damage consists both of the pavement repair costs incurred by the government and increased operating costs for vehicles attributable to bumpier roads. However, if the government steps in to repair roads once they reach a predetermined state of deterioration, then as a rough rule of thumb, the total external cost of road damage can be measured by average annual spending on maintaining the road network.24

A complication is that road damage is jointly caused by vehicle traffic and weather, such as ice creating and exacerbating holes and cracks in the pavement, which are further enlarged by vehicle traffic. Empirical work for apportioning damage to trucks versus weather is sparse. Depending on road strength (e.g., thickness), Paterson (1987, p. 372) suggests that vehicles cause 40–90 percent of damage in warm, dry, or subhumid climates; 20–80 percent in arid, freezing climates; and 10–60 percent in moist, freezing climates.25 Here trucks are assumed to account for 50 percent of the damage in all countries.26

IRF (2009, Table 8.2) provides separately spending on road maintenance and capacity investments aggregated over all levels of government for 2007 (or the latest available) for 74 countries. For other countries, IRF (2009) provides total highway spending, but not the maintenance-to-investment decomposition; therefore, this breakdown is inferred from a similar country in the same region for which the breakdown is available. For the remaining 10 countries for which no spending data are available, maintenance expenditure per truck-kilometer is assumed to be the same as a similar country in the same region. Scaling by the share of trucks (as opposed to climate) in total damage gives the damage attributable to trucks by country.


Congestion costs are estimated using extrapolations of travel delays from a city-level database to the country level (in the absence of direct data on these delays) and travel time valuations from the literature. Accident costs are assessed by making assumptions about what portion of road fatalities in different countries reflects risks that motorists do not take into account, and making some upward adjustments to allow for other components of accident risk (property damage, medical costs, and nonfatal injuries). Both congestion and accident costs are sizable (and likely understated); in some cases, accident costs exceed congestion costs. Road damage is also estimated by attributing a portion of road maintenance expenditures to trucks, though these costs are modest in relative terms. A few extra steps are needed to calculate corrective motor fuel taxes; see Annex 5.7 for details.

All of the cost estimates are rudimentary and there will be ample scope for reforming them in the future as data, such as on travel delays, become more widely available and analytical work helps to resolve some of the uncertainties (e.g., about the VOT in low-income countries, or the safety risks that one driver poses to other road users). In the meantime, however, these cost estimates enable a first-pass estimate of corrective motor fuel taxes that can be studied and might serve as a useful starting point for discussions about tax reform.

Annex 5.1. Measuring Congestion Costs: Some Technicalities

The total hourly costs (TC) of congestion to passengers in vehicles driving along a one-kilometer lane-segment of a highway can be expressed as


in which V denotes traffic volume or flow—the number of cars that pass along the kilometer-long stretch per hour (the implications of other vehicles on the road are discussed below). Tf is travel time per kilometer when traffic is freely flowing, and T (which exceeds Tf) is the actual travel time, an increasing function of the traffic volume (speeds fall with less road space between vehicles). The component o is vehicle occupancy, or average number of passengers per vehicle. The total travel delay from congestion for all passengers is therefore V × (T—Tf) × o, where TTf is the average delay per vehicle-kilometer. Multiplying total travel delay by the value of travel time (VOT) expresses delays as a monetary cost.

Dividing TC by traffic volume gives the average cost of congestion (AC) per vehicle-kilometer:


AC is the cost borne by individual motorists that, on average, they should take into account when deciding how much to drive.

Differentiating TC with respect to V gives the added congestion cost to all road users from an extra vehicle-kilometer:


This added cost includes the average cost (taken into account by the driver), as just described. It also includes the cost to occupants of other vehicles, which is not taken into account by the driver. The latter is the delay to other vehicles, (dT/dV) × V, times the average number of people in other vehicles, times the VOT to express costs in monetary units.

Suppose, as discussed in the main text, that travel delay can be approximated by a power function of traffic volume, that is,


in which α and β are constants. The constant α reflects factors like road capacity, and β reflects the rate at which additional traffic diminishes travel speeds. Differentiating this expression by V gives dT/dV = αβVβ-1. Then using (5.4) gives the following:


The delay to other vehicles, (dT/dV) × V, simply the product of average delay and the scalar β. As discussed in the main text, empirical studies suggest a value for β of between about 2.5 and 5.0 for congested roads.

If speed data are available, average delay can be estimated using the following equation:


in which S and Sf are the actual and the free-flow travel speeds (kilometers/hour).

Annex 5.2. Cities Comprising City-Level Database

Cities covered in the city-level database, which is used to obtain statistical relationships between travel delays and various transportation indicators, are listed in Annex Table 5.2.1.

Annex Table 5.2.1

Cities in the City-Level Database

(Used to Extrapolate Congestion Costs)

article image
article image
Source: See main text.Note: Excludes 10 cities from the original database that were dropped because of insufficient data.

Annex 5.3. Results from Statistical Methods Used to Relate Travel Delay to Travel Indicators

As discussed in the main text, statistical regressions were used to estimate the contribution of various factors to explaining travel delays across the 90 cities in the database. To obtain the best statistical fit (i.e., to reduce the noise from outlying observations or extreme values), average delay and the four explanatory indicators are expressed in natural logarithm form in the regression and second powers of these variables are included. (Both of these are standard statistical procedures.) The regression results are presented in Annex Table 5.3.1.

Annex Table 5.3.1

Regression Results for City-Level Average Delay

article image
Source: See main text.Note: *, **, and *** indicate significance at the 10 percent, 5 percent, and 1 percent levels, respectively.

Interpreting these coefficients is less of a concern than the statistical fit (which is reasonably good) because the coefficients are used for prediction rather than for establishing causal relationships. In fact, the explanatory variables such as road length per car and cars per capita are likely to be simultaneously determined with traffic conditions such as average delay (the dependent variable), which confounds the interpretation of the coefficients.27

As noted in the main text, car-kilometers driven is not available at the country level for 81 countries. To fill in this gap, countries are grouped by region (Europe, Oceania, Africa, and so on) and statistical regressions are used to estimate a relationship for each regional grouping between car-kilometers (for countries for which these data are available) and four explanatory variables that are available for all countries: per capita income, urban population density, vehicle ownership, and road density (using data from IRF, 2009, and World Bank, 2013). Using this relationship, and the explanatory variables, kilometers driven per car are then derived for countries for which direct data are missing.

The natural logarithm of kilometers driven per car and the four explanatory indicators (of the 69 countries that have complete data) was taken and the second and third power of the log explanatory variables were included to add more flexibility. The regression results are shown in Annex Table 5.3.2, though again, because the equation is used for prediction, the interpretation of the estimated coefficients is not especially of concern.

Annex 5.4. Accounting for Delays to All Vehicle Occupants

This annex presents illustrative calculations to show how congestion cost estimates change when the mix of cars, buses, trucks, and two-wheel motorized vehicles is taken into account (the formulas in Annex 5.1 assume cars are the only vehicles).

Following from equation (5.1), the total costs of travel delays to all road users, when accounting for different vehicle types, is given by:

Annex Table 5.3.2

Regression Results for Kilometers Driven per Car

article image
Source: See main text.Note: *, **, and *** indicate significance at the 10 percent, 5 percent, and 1 percent levels, respectively.

Subscript i is used to denote a particular type of vehicle: i = car, bus, truck, or two-wheeler. For simplicity, congestion is assumed to increase delay for all vehicles by the same absolute amount.

Differentiating equation (5.7) with respect to Vcar, and using the definition of AC from equation (5.2), gives the following:


Comparing equations (5.3) and (5.8), the ratio of the cost imposed on other vehicle occupants when there is a mix of vehicles as opposed to just cars is given by


in which Vi/V is the share of vehicle i in total kilometers driven by all vehicles.

For the calculations in the remainder of this annex, the occupancy of trucks and two-wheelers is assumed to be one, and (based approximately on Parry and Small, 2009, for cities in the United Kingdom and the United States) that for buses is assumed to be 20. The VOT for two-wheelers and bus riders is assumed to be the same as for car occupants. For freight travel by trucks, the VOT should include the employer wage (the market wage plus employer payroll taxes) to reflect the per hour costs of labor time lost from congestion. Given that the VOT for car travel is 60 percent of the market wage, this implies VOTtruck/VOTcar = 1.67.

The last column of Annex Table 5.4.1 shows, based on equation (5.9), congestion costs with different scenarios for the vehicle fleet mix relative to congestion costs when cars are the only vehicles on the road.

Annex Table 5.4.1

Ratio of Congestion Cost with Multiple Vehicles Relative to Costs when Cars are the Only Vehicle

article image
Source: See text of Annex 5.4.

If the only other vehicles are trucks and two-wheelers, there is relatively little difference in the results: in Annex Table 5.4.1, congestion costs are increased 1 percent when trucks account for 20 percent of the fleet and are reduced 7 percent when two-wheelers account for 20 percent of the fleet (with, in both cases, cars accounting for the remaining 80 percent). However, when buses account for 10 percent of the vehicle fleet congestion costs more than double, and when they account for 20 percent, costs more than triple. A car driver imposes considerably higher costs on others when the average number of vehicle occupants is higher, resulting from a significant share of high-occupancy buses on the roads. Congestion costs and motor fuel taxes may therefore be substantially understated in countries in which buses account for a significant share of vehicle traffic in urban centers.

Annex 5.5. Assessment of Congestion Costs from Country-Level Data: The United States and the United Kingdom

This annex explains the supplementary estimation of delays at the country level for the United States and the United Kingdom, mentioned in the main text. Estimated delays are for 2008 (as a close approximation to 2010) and costs are expressed in 2010 U.S. dollars.

The United States

For the United States, the Texas Transportation Institute (TTI) compiles high-quality data on travel delays for 449 urban centers categorized by population size into very large, large, medium, and small cities (Schrank, Lomax, and Eisele, 2011).

For the 101 largest cities, speed data are collected remotely by a private company for different times of the day for each link within the urban road network. For the other 348 smaller urban centers (which account for 15 percent of nationwide travel delays), speed is derived from estimated speed/traffic volume relationships. Schrank, Lomax, and Eisele (2011) use traffic volume data from the Highway Performance Monitoring System, an inventory maintained by the Federal Highway Administration for all roadway segments in the United States.

The TTI report for 2008 was used to derive the nationwide congestion delay on others. For each urban region in the TTI sample, total annual hours of delay to passengers in cars is divided by total annual vehicle-kilometers driven by cars to give the average hourly delay per car-kilometer. Delays at the regional level are weighted by the share of car-kilometers in nationwide kilometers and then aggregated to obtain a nationwide average measure of delay.

The United Kingdom

For the United Kingdom, travel data for 2008 were obtained from the U.K. Department for Transport (DFT), which compiles official statistics on the British transport system. Because DFT does not provide annual hours of travel delays at the city level, travel delays were generated by comparing average vehicle speed during peak times with the free flow speed, both of which can be obtained from the DFT statistics.28

For each of the U.K. localities, the average travel time per kilometer was calculated along with the free flow average travel time per kilometer, using the average travel speed during morning peak (7 am to 10 am) and the free flow speed.

Annual car-kilometers within each locality were then multiplied by the share of car-kilometers occurring during the morning peak period. The total annual hours of delay was then obtained by multiplying the extra travel time per kilometer during morning peak time by 2 to account for the evening peak (4 pm to 7 pm) which is assumed to experience the same traffic congestion.29

Next, the total annual hours of delay were divided by total annual vehicle-kilometers driven by cars for each locality to derive the average hourly delay per vehicle-kilometer, which was then converted into passenger delays assuming an average vehicle occupancy of 1.6. Average delay at the nationwide level is a weighted average of that at the locality level, with weights equal to the shares in nationwide car-kilometers.

Delays to others per car-kilometer are about twice as high for the United Kingdom as for the United States, which seems roughly plausible, given that a much greater share of nationwide driving occurs under congested conditions in the United Kingdom.

Annex 5.6. Estimating the Ratio of Other Accident Costs to Fatality Costs: Country Case Studies

As mentioned in the main text, external accident costs for different countries are obtained by scaling up estimates of external fatality costs by the ratio of other costs to external fatality costs. This ratio—which is based on several country case studies—is attained as follows:

First, using data compiled by Herrnstadt, Parry, and Siikamäki (2013) for Finland, Sweden, the United Kingdom, and the United States and by Parry and Strand (2012) for Chile, comprehensive estimates of external accident costs were made for these five countries for 2010 or the latest possible. In these calculations, external fatalities were monetized using mortality values discussed in Chapter 4. Other costs were valued using a combination of local data on the average (personal, medical, and property damage) costs associated with accidents of different severity, and in some cases, extrapolations of these costs from U.S. data.30 Approximately 85 percent of medical costs are assumed to be external (borne by third parties) for all fatal and nonfatal and internal and external injuries, and 50 percent of property damage costs (for all accidents) are external.

From these studies, five point estimates for the ratio of other external costs (medical, property damage, and nonfatal injury costs) to external fatality costs were obtained. This ratio tends to decline as the relative importance of pedestrian and other external deaths in total road deaths rises (the numerator in the ratio falls and the denominator rises). This ratio is 2.9 in the United States (where 23 percent of deaths are external) and only 0.16 in Chile (where 54 percent of deaths are external). A power function that best fits these five data points relating this cost ratio to the share of external fatalities in total fatalities was estimated.31 This relationship was then used to extrapolate the other-external-cost-to-fatality-external-cost ratio for other countries depending on their share of external fatalities in total fatalities.

Annex 5.7. Miscellaneous Data and Procedures for Calculating Corrective Taxes

The remaining data and assumptions needed to implement the corrective fuel tax formula as set forth in Chapter 3 are discussed in this annex. These issues deal with the use of diesel fuel by both cars and trucks; the breakdown of fuel price responses; and the conversion of road damage, accident, local pollution, and congestion costs into corresponding components of corrective fuel taxes. To make this conversion, fuel efficiency is needed to convert any road damage, accident, local pollution, or congestion costs expressed per vehicle-kilometer driven into a cost per liter of fuel. However, given the difficulty of accurately measuring fuel efficiency for most countries (see below), these costs are directly expressed per liter insofar as possible, to avoid the need for this data.

Diesel use by different vehicle types: External costs for cars are used to calculate corrective taxes on gasoline. However, diesel fuel is used by both cars and trucks and, given the practical difficulty of differentiating the diesel tax according to vehicle use, a weighted average of external costs for cars and trucks should be used in the corrective diesel tax formula, based on their respective shares in diesel fuel consumption. The breakout of diesel fuel use by cars versus trucks is available for a limited number of countries and for other countries was taken from regional average figures.32

Breakdown of fuel price responses: An important piece of data is the fraction of the fuel demand response that comes from reduced driving (as opposed to the remaining fraction that comes from fuel efficiency improvements). For cars, this fraction is assumed to be 0.5 for all countries.33 For diesel fuel used by trucks (where the high power requirements necessary to move freight limit opportunities for improving fuel efficiency through, for example, reducing vehicle size and weight) this fraction is assumed to be 0.6 (Parry, 2008).

Road damage: The estimation procedures outlined in the main text yield total external costs for road damage. These costs are divided by total diesel fuel consumption for trucks to obtain costs per liter, which are then multiplied by 0.6 to account for the portion of the fuel response that comes from reduced kilometers driven.

Accidents: The estimation procedures also yield total external costs for traffic accidents. However, expressing them per liter of fuel is more involved because external costs per vehicle-kilometer are assumed to be the same for cars and trucks, implying external costs per liter of fuel will be larger for cars than for trucks given that cars travel farther on a liter of fuel, and larger for diesel fuel cars than gasoline cars (because diesel cars are more fuel efficient). Truck fuel efficiency is assumed to be one-third that for gasoline cars (Parry, 2008), and in turn, diesel cars are assumed to be 20 percent more fuel efficient than their gasoline counterparts.

External accident costs per liter of gasoline can then be obtained by dividing total accident costs for all vehicles by a weighted sum of fuel use by gasoline cars, diesel cars, and trucks (fuel use data are discussed in the Annex 6.1); the weights are fuel efficiency of other vehicles relative to that for gasoline cars (1.2 and 0.33, respectively, for diesel cars and trucks). In turn, costs per liter for diesel cars and trucks are the external costs per liter for gasoline cars multiplied by the same weights. In applying the costs to the corrective fuel tax formula, they are again multiplied by the portion (0.5 or 0.6) of the fuel price response that comes from reduced driving as opposed to fuel efficiency improvements.

Local pollution: Local pollution damage is estimated on a per liter basis. The scaling factor, however, depends on how emissions are regulated. In countries such as the United States, where emissions are regulated on a per kilometer (or per mile) basis and approximately maintained throughout the vehicle’s life, roughly speaking emissions vary only with kilometers driven, not fuel efficiency, and therefore need to be multiplied by the driving fraction of the fuel price response.34 In countries with less effective regulation, emission might be appropriately proportional to fuel use. The calculations in this analysis apply a scaling factor of 0.5 (gasoline vehicles) or 0.6 (diesel vehicles) for Australia, Canada, China, European countries, New Zealand, Singapore, and the United States, and 1.0 (no adjustment) for all other countries. More refined assumptions would not have that much effect on the corrective fuel tax estimates given the relatively large size of congestion and accident costs (see Chapter 6).

Congestion: Congestion costs are estimated on a per kilometer basis and therefore need to be multiplied by fuel efficiency (see below) to express them in per liter terms (after scaling by the driving fraction of the fuel price response).

One complication is that driving on congested roads (mostly by people commuting to work) is generally less sensitive to fuel prices than driving on uncongested roads. This fact reduces the congestion benefits from higher fuel taxes. Based on evidence of the relative price responsiveness of driving under congested and noncongested conditions, Parry and Small (2005) recommend scaling back congestion costs by a third in computing corrective fuel taxes; the same procedure is followed here.

Fuel efficiency: Fuel efficiency (of vehicles in use on the road) could be obtained by dividing data on vehicle-kilometers driven by fuel use. However, because the reliability of the vehicle-kilometer data varies across countries (being generally less accurate for developing countries), fuel efficiency is based instead on a plausible assumption for different regions, and applied to all countries in the region. For example, based on estimates in Parry and Small (2005) for the United States and the United Kingdom, fuel efficiency for gasoline vehicles is assumed to be 10.5 kilometers/liter (25 miles/gallon) in North America and 14.5 kilometers/ liter (35 miles/gallon) for higher-income European countries and Japan.35 Fuel efficiency for diesel cars and trucks is then derived using the above ratios (1.2 and 0.33, respectively). Other assumptions would moderately affect the contribution of congestion costs to corrective taxes.

Finally, to simplify the computation of corrective fuel taxes, it is assumed that fuel efficiency in each country remains fixed at its current level, rather than increasing in response to higher fuel prices. This assumption leads again to some understatement of the corrective fuel tax because, per liter of fuel reduction, the reduction in vehicle-kilometers driven (and hence congestion and accidents) is greater for a more fuel-efficient vehicle; see equation (3.1).36


  • Abrantes, P.A.L., and M.R. Wardman, 2011, “Meta-Analysis of UK Values of Travel Time: An Update,Transportation Research Part A: Policy and Practice, Vol. 45, No. 1, pp. 117.

    • Search Google Scholar
    • Export Citation
  • Becker, Gary S., 1965, “A Theory of the Allocation of Time,The Economic Journal, Vol. 75, pp. 493511.

  • Bureau of Transportation Statistics (BTS), 2012, National Transportation Statistics (Washington: Bureau of Transportation Statistics, US Department of Transportation).

    • Search Google Scholar
    • Export Citation
  • Commissariart Général du Plan, 2001, Transports: Choix des Investissements et Coût des Nuisances [Transportation: Choice of Investments and the Cost of Nuisances] (Paris).

    • Search Google Scholar
    • Export Citation
  • Duranton, Gilles, and Matthew A. Turner, 2011, “The Fundamental Law of Road Congestion: Evidence from US Cities,American Economic Review, Vol. 101, pp. 261652.

    • Search Google Scholar
    • Export Citation
  • Edlin, Aaron S., and Pinar Karaca-Mandic, 2006, “The Accident Externality from Driving,Journal of Political Economy, Vol. 114, pp. 93155.

    • Search Google Scholar
    • Export Citation
  • Eliasson, Jonas, 2006, “Forecasting Travel Time Variability,Proceedings of the European Transport Conference (Henley-in-Arden, U.K.: Association for European Transport).

    • Search Google Scholar
    • Export Citation
  • Fosgerau, M., K. Hjorth, C. Brems, and D. Fukuda, 2008, Travel Time Variability: Definition and Valuation (Lyngby, Denmark: DTU Transport).

    • Search Google Scholar
    • Export Citation
  • Fridstrøm, Lasse, Jan Ifver, Siv Ingebrigtsen, Risto Kulmala, and Lars Krogsgård Thomsen, 1995, “Measuring the Contribution of Randomness, Exposure, Weather, and Daylight to the Variation in Road Accident Counts,Accident Analysis and Prevention, Vol. 27, pp. 120.

    • Search Google Scholar
    • Export Citation
  • Greenwood, Ian D., and Christopher R. Bennett, 1996, “The Effects of Traffic Congestion on Fuel Consumption,Road and Transport Research, Vol. 5, pp. 1831.

    • Search Google Scholar
    • Export Citation
  • Herrnstadt, Evan, Ian W.H. Parry, and Juha Siikamäki, 2013, “Do Alcohol Taxes in Europe and the US Rightly Correct for Externalities?International Tax and Public Finance, published online September 25.

    • Search Google Scholar
    • Export Citation
  • International Council on Clean Transportation (ICCT), 2010, Global Transportation Roadmap Model (Washington: International Council on Clean Transportation).

    • Search Google Scholar
    • Export Citation
  • International Labor Organization (ILO), 2012, Global Wage Database 2012 (Geneva: International Labor Organization).—en/index.htm.

    • Search Google Scholar
    • Export Citation
  • International Road Federation (IRF), 2009, World Road Statistics 2009 (Geneva: International Road Federation).

  • International Road Federation (IRF), 2012, World Road Statistics 2012 (Geneva: International Road Federation).

  • Kopits, Elizabeth, and Maureen Cropper, 2008, “Why Have Traffic Fatalities Declined in Industrialized Countries? Implications for Pedestrians and Vehicle Occupants,Journal of Transport Economics and Policy, Vol. 42, pp. 12954.

    • Search Google Scholar
    • Export Citation
  • Lindberg, Gunnar, 2001, “Traffic Insurance and Accident Externality Charges,Journal of Transport Economics and Policy, Vol. 35, pp. 399416.

    • Search Google Scholar
    • Export Citation
  • Lindsey, Robin, 2010, “Dedicated Lanes, Tolls and Its Technology,” in The Future of Interurban Passenger Transport (Paris: Organization for Economic Cooperation and Development).

    • Search Google Scholar
    • Export Citation
  • Mackie, P.J., M. Wardman, A.S. Fowkes, G. Whelan, J. Nellthorp, and J. Bates, 2003, “Values of Travel Time Savings in the UK: Summary Report” (Leeds, U.K.: Institute for Transport Studies, University of Leeds).

    • Search Google Scholar
    • Export Citation
  • Newbery, David M., 1987, “Road User Charges in Britain,Economic Journal, Vol. 98, No. 390 (Supplement), pp. 16176.

  • Newbery, David M., 1990, “Pricing and Congestion: Economic Principles Relevant to Pricing Roads,Oxford Review of Economic Policy, Vol. 6, pp. 2238.

    • Search Google Scholar
    • Export Citation
  • Parry, Ian W.H., 2004, “Comparing Alternative Policies to Reduce Traffic Accidents,Journal of Urban Economics, Vol. 56, pp. 34668.

    • Search Google Scholar
    • Export Citation
  • Parry, Ian W.H., 2008, “How Should Heavy-Duty Trucks be Taxed?Journal of Urban Economics, Vol. 63, pp. 65168.

  • Parry, Ian W.H., and Kenneth A. Small, 2005, “Does Britain or the United States Have the Right Gasoline Tax?American Economic Review, Vol. 95, No. 4, pp. 127689.

    • Search Google Scholar
    • Export Citation
  • Parry, Ian W.H., 2009, “Should Urban Transit Subsidies Be Reduced?American Economic Review, Vol. 99, pp. 70024.

  • Parry, Ian W.H., and Jon Strand, 2012, “International Fuel Tax Assessment: An Application to Chile,Environment and Development Economics, Vol. 17, pp. 12744.

    • Search Google Scholar
    • Export Citation
  • Paterson, William D.O., 1987, Road Deterioration and Maintenance Effects—Models for Planning and Management, World Bank Highway Design and Maintenance Standards Series (Baltimore, Maryland: Johns Hopkins University Press).

    • Search Google Scholar
    • Export Citation
  • Peer, S., C.C. Koopmans, and E.T. Verhoef, 2012, “Prediction of Travel Time Variability for Cost-Benefit Analysis,Transportation Research Part A: Policy and Practice, Vol. 46, pp. 7990.

    • Search Google Scholar
    • Export Citation
  • Schrank, David, Tim Lomax, and Bill Eisele, 2011, 2011 Urban Mobility Report (College Station, Texas: Texas Transportation Institute, Texas A&M University.

    • Search Google Scholar
    • Export Citation
  • Small, Kenneth A., 1992, Urban Transportation Economics, In Fundamentals of Pure and Applied Economics Series, vol. 51 (Chur, Switzerland: Harwood Academic Press).

    • Search Google Scholar
    • Export Citation
  • Small, Kenneth A., and Jose A. Gómez-Ibáñez, 1998, “Road Pricing for Congestion Management: The Transition from Theory to Policy,” in Road Pricing, Traffic Congestion and the Environment: Issue of Efficiency and Social Feasibility, edited by K.J. Button and E.T. Verhoef (Cheltenham, U.K.: Edward Elgar) pp. 21346.

    • Search Google Scholar
    • Export Citation
  • Small, Kenneth A., and Kurt Van Dender, 2006, “Fuel Efficiency and Motor Vehicle Travel: The Declining Rebound Effect,Energy Journal, Vol. 28, No. 1, pp. 2552.

    • Search Google Scholar
    • Export Citation
  • Small, Kenneth A., and Erik Verhoef, 2007, The Economics of Urban Transportation (New York: Routledge).

  • Small, Kenneth A., Clifford Winston, and Carol A. Evans, 1989, Road Work (Washington: Brookings Institution).

  • Transport Canada, 1994, Guide to Benefit-Cost Analysis in Transport Canada (Ottawa, Ontario: Transport Canada).

  • Transport Canada, 2006, Costs of Non-Recurrent Congestion in Canada. Final Report. TP 14664E (Ottawa, Ontario: Transport Canada).

  • Transportation Research Board, 2010, Highway Capacity Manual 2010 (Washington: National Academies).

    • Search Google Scholar
    • Export Citation
  • Traynor, Thomas L., 1994, “The Effects of Varying Safety Conditions on the External Costs of Driving,Eastern Economic Journal, Vol. 20, pp. 4560.

    • Search Google Scholar
    • Export Citation
  • United States Department of Transportation, 1997, The Value of Travel Time: Departmental Guidance for Conducting Economic Valuations (Washington: US Department of Transportation).

    • Search Google Scholar
    • Export Citation
  • United States Federal Highway Administration (US FHWA), 1997, 1997 Federal Highway Cost Allocation Study (Washington: Federal Highway Administration, US Department of Transportation).

    • Search Google Scholar
    • Export Citation
  • United States Federal Highway Administration (US FHWA), 2005, Crash Cost Estimates by Maximum Police-Reported Injury Severity Within Selected Crash Geometries, FHWA-HRT-05—051 (Washington: Federal Highway Administration, US Department of Transportation).

    • Search Google Scholar
    • Export Citation
  • Wardman, Mark, 1998, “The Value of Travel Time: A Review of British Evidence,Journal of Transport Economics and Policy, Vol. 32, pp. 285316.

    • Search Google Scholar
    • Export Citation
  • Wardman, Mark, 2001, “A Review of British Evidence on Time and Service Quality Valuations,Transportation Research E, Vol. 37, pp. 10728.

    • Search Google Scholar
    • Export Citation
  • Waters, William G. II, 1996, “Values of Time Savings in Road Transport Project Evaluation,” in World Transport Research: Proceedings of 7th World Conference on Transport Research, vol. 3., edited by David Hensher, J. King, and T. Oum (Oxford, U.K.: Pergamon) pp. 21323.

    • Search Google Scholar
    • Export Citation
  • World Bank, 2013, World Development Indicators Database (Washington: World Bank).

  • World Health Organization (WHO), 2013, Global Health Observatory Data Repository. (Geneva: World Health Organization). = en.

    • Search Google Scholar
    • Export Citation

The database was developed by the International Association of Public Transport (UITP) and the Institute for Sustainability and Technology (ISTP) in 2001.


Most of the speed data are calculated using traffic counts and assumptions about how speed varies with traffic volume.


A study for Canada, for example, suggests that nonrecurrent congestion costs could be as large as those for recurrent congestion (Transport Canada, 2006).


These free flow speeds (which are not available in the data) are assumed to be 57 kilometers/hour (35 miles/hour) or 65 kilometers/hour (40 miles/hour), according to whether cities have relatively high or relatively low road density per urban hectare (in fact, for some cities, the observed travel speeds are close to the free flow speeds). These assumptions are roughly in line with those in Parry and Small (2009).


The most recent data are for 2007, which is assumed to provide a reasonable approximation for delays in 2010.


In a handful of cases for which the results looked especially questionable, average delays per kilometer were extrapolated from other countries. For example, delays for Bangladesh and Kazakhstan were extrapolated from India and Russia, respectively, and included an adjustment for differences in urbanization rates between countries.


This assumption is consistent with the Bureau of Public Roads formula, the traditional method for predicting vehicle speed as a function of the volume-to-capacity ratio. See Small and Verhoef (2007, pp. 69–83) and Small (1992, pp. 70–71) for further discussion. Obviously the above approach is highly simplified—speed/volume relationships may vary considerably with the characteristics of specific roads (e.g., speed limits, frequency of stop lights and sharp bends) and across different times of day. But the assumed value for β seems to be a reasonable rule of thumb for representing average travel conditions in urban areas.


This is slightly higher than the average vehicle occupancy rates for London, Los Angeles, and Washington calculated in Parry and Small (2009).


Much evidence, at least from advanced countries, suggests that the VOT increases approximately in proportion to income, which backs up this assumption (Small and Verhoef, 2007, p. 52). Abrantes and Wardman (2011) determine that a 10 percent increase in income increases the VOT by 9 percent. It might be argued that the VOT should be adjusted upward in countries with relatively low vehicle ownership rates, where ownership is skewed toward higher wage groups. No adjustments are made, however, partly because of data limitations. But the issue is not clear cut either—conceivably, higher-income people (at least those living in more expensive housing closer to downtown areas) drive less under congested conditions than do other motorists.


There are data gaps for six countries in ILO (2012). For these cases, wages are proxied using GDP per capita. Ideally, urban wage rates (adjusted downward for differences compensating for higher living costs) would be used in preference to nationwide wages, but a comprehensive, international data set is not available.


There are some nuances. The relative differences between developed and developing countries are a bit more pronounced for the VOT because relative wages across countries are compared whereas Figure 4.2 compares relative income raised to the power 0.8. There are also some differences even among similar-income countries. For example, the United States has a higher mortality valuation than Australia but slightly lower VOT, reflecting the depressing effect on U.S. wages of relatively high labor force participation among migrants and secondary family workers, and relatively little influence of labor unions or labor market regulations on inflating wages.


In many cases the bus share is very low (e.g., about 1 percent or less of vehicle-kilometers traveled in Washington, Los Angeles, and London—see Parry and Small, 2009).


Other costs from traffic accidents, such as those from traffic holdups, police and fire services, insurance administration, and legal costs, are beyond the scope of this chapter. According to some studies, they appear to be modest relative to other costs (e.g., US FHWA, 2005; Parry, 2004, Table 2). That might seem surprising for traffic holdups given that some accidents cause severe traffic disruptions, but these accidents constitute only a small share of total accidents. Productivity losses are taken into account in the monetary values assigned to different types of injuries.


However, once on the road, drivers likely take care to lower the risks of hitting pedestrians and cyclists. Because observed injury data reflect this likelihood, it is taken into account in the corrective tax estimates.


Motorists may lack an accurate sense of risks to themselves but, in the absence of evidence to the contrary, it seems reasonable to assume that the average motorist does not systematically understate or overstate these risks—and even if there were such evidence, information campaigns to better educate drivers might be a better response than corrective fuel taxes.


For example, Edlin and Karaca-Mandic (2006) find that extra driving substantially increases average insurance costs per kilometer driven, suggesting higher per kilometer property damage costs (though how other costs, like fatality risk, change is not clear). However, studies by Lindberg (2001), Traynor (1994), and Fridstrøm and others (1995) suggest that extra driving may have only limited effects, and possibly even a negative effect, on severity-adjusted accident risk.


Premiums may also vary moderately with an individual’s stated annual driving, which also provides some, albeit very weak, link between extra driving and property damage (in the form of greater premiums) paid by drivers. And to the extent that insurance companies have market power, motorists may be taxed, in effect, for risks of property damage.


In 2010, the crash frequency per kilometer driven in the United States for light-duty vehicles was almost four times that for trucks (BTS, 2012, Tables 2.21 and 2.23).


For example, fatalities in India were 133,938 for 2010 according to IRF (2012), but were estimated to be 231,027 in 2010 by WHO (2013).


In principle, it might seem that a higher value should be used for traffic-related deaths, given that the average age of someone dying in a road accident is lower than for the average person dying from pollution-related illness (Small and Verhoef, 2007, p. 101). However, for reasons discussed in Box 4.3 of Chapter 4, an adjustment is not made.


Property damage accounts for 42 percent of other external costs, nonfatal injuries 38 percent, and medical costs 20 percent, based on a simple average across the studies.


In India, for example, there are 40 external deaths per billion vehicle-kilometers compared with 2 in the United States.


Road damage increases approximately in proportion to the third power of a vehicle’s axle weight (Small, Winston, and Evans, 1989), though there is considerable variation across road surfaces.


If roads are repaired more frequently, government resource costs are higher but there is less deterioration of vehicle operating costs, and vice versa. See Newbery (1987) for a more precise discussion.


See also Newbery (1987) for similar findings.


A more accurate calculation (but one that would have little impact on overall corrective diesel fuel tax estimates) would involve classifying countries by climate zone and, better still, average road strength, and applying different assumptions about the proportion of damage attributable to trucks.


For example, the negative sign for kilometers per car suggests, perhaps, that extra traffic creates pressure or incentives for road investment (Duranton and Turner, 2011) or that bad traffic conditions discourage driving.


The data used are from, specifically data sets CGN0201, SPE0104, TRA8901, and TRA0307.


According to DFT traffic distribution data (TRA0307), the shares of morning and evening peak kilometers in total kilometers driven are 0.21 and 0.22, respectively, which are very close.


Detailed documentation of data sources and estimation procedures are provided in the above references. The breakdown of fatalities by driver, passenger of drivers, other vehicle occupant, pedestrian, and cyclist is available from the data sources, and this breakdown is assumed to be the same for nonfatal injuries. Herrnstadt, Parry, and Siikamäki (2013) focus only on alcohol-related accidents; their data were modified to include data for all traffic accidents.


Specifically, the cost ratio is predicted by the equation 0.049 x-2.56, in which x is the share of external fatalities in total fatalities.


The data source is ICCT (2010). For example, cars account for about 11 percent of road diesel at the global level, and 32 percent in the European Union.


See Small and Van Dender (2006) and the review of other studies in Parry and Small (2005). In practice this fraction will vary across countries; for example, it might be higher in countries with readily available alternatives to car use (which increases the responsiveness of driving to fuel prices) and in countries with binding fuel efficiency regulations (which reduces the responsiveness of fuel efficiency to fuel prices). However, no international data on which country-specific assumptions can be based are available.


In some cases, emissions standards are defined with respect to engine capacity (e.g., in European Union countries as well as in other countries adopting European Union standards). In this analysis, some fuel efficiency improvements, such as reducing vehicle weight, will affect emissions but others, such as more efficient engines, will not.


These figures make some adjustment for recent increases in fuel efficiency. Other assumptions are Central and South America and Eurasia, 10.5 kilometers/liter; lower-income Europe and Asia, 12.5 kilometers/liter; and Middle East, 8.5 kilometers/liter.


The understatement is not huge, however. For example, based on assumptions here and in Annex 6.1, even a 50 percent increase in gasoline prices would increase fuel efficiency by 12.5 percent (see also Small and Van Dender, 2006).

From Principle to Practice
  • View in gallery

    Value of Travel Time, Selected Countries, 2010

  • View in gallery

    Congestion Costs Imposed on Others per Car-Kilometer, Selected Countries, 2010

  • View in gallery

    Congestion Costs Imposed on Others per Car-Kilometer, All Countries, 2010

  • View in gallery

    External Accident Costs per Vehicle-Kilometer, Selected Countries, 2010

  • View in gallery

    External Accident Costs per Kilometer Driven, All Countries, 2010