Integrating Accident and Travel Delay Externalities in an Urban Speed Reduction Context

Abstract

Accident externality costs remain controversial in terms of their costing and valuation. Much of the literature on accident and travel delay externalities treats each source as mutually exclusive and additive, yet common sense suggests that interdependencies prevail. One example of this is the recognition that accident externalities are not independent of travel delays, and hence travel time savings and losses are influenced by policy designed to reduce the risk of exposure to accidents. Reduced maximum speed limit restrictions also add costs in terms of loss of travel time (and increased speed limits produce travel time benefits). Also, lowered speed limits may lead to more drivers risking exceeding the speed limit because of perceived time loss, thereby exacerbating the potential for accidents. The paper takes a close look at the empirical relationship between accident and travel delay externalities in an urban setting, accounting for the risk‐compensating behaviour under conditions of greater accident risk. Recognizing that levels of risk in an urban setting are broadly a function of traffic densities and that the latter can be approximated by the mix of free flow and non‐free flow travel time (for a given total travel time), an aggregate marginal externality cost function is used to quantify empirically the input elements in the context of a driver’s choice between a free and a tolled route in Sydney, Australia. This discrete choice context is sufficient, given an externally established relationship between speed and traffic density, to quantify the marginal externality accident and travel time delay costs. It is shown what additional externality has to be factored into the accident costs to recognize the other sources of externality typically ignored in accident costing and speed restriction studies.

Acknowledgements

Ken Ogden, General Manager (Public Relations), Royal Automobile Club of Victoria, is thanked for getting the author interested in this theme and in providing some background material to set the context for the paper. The opportunity to see a draft of Seiji Steimetz’s doctoral thesis (University of California Irvine), where he suggested that my research on separating out free flow and delay time provides a way forward in valuing marginal reductions in factors such as risk and effort, also motivated the emphasis herein. The comments of two referees have materially improved the paper.

Notes

1. A referee pointed out that speed restrictions are likely to reduce observed accident rates, but may still increase accident costs due to the increased defensive efforts that more dense traffic conditions require.

2. Effort is a complex construct in need of careful definition. The paper focuses on effort linked to self‐protection. This can include (at least) two observed types: (1) a higher density demands more effort from the driver to avoid collisions with the increasing number of cars sharing the infrastructure; and (2) driving at higher speeds also requires another type of effort. These classes of effort may entail different sets of costs.

3. Fred Mannering, Purdue University, personal correspondence (8 October 2004), states that the cost and value of speed limits is an important topic, specifically the trade‐off drivers make between speed and safety, and the manner in which society sets speed limits. On a somewhat related note, he states that research with Cliff Winston suggests that drivers compensate for safety features in cars by driving more aggressively. That is, they are maintaining their risk level and extracting the benefits of vehicle safety devices via increased mobility (speed). It is also possible that drivers themselves are lowering their risk of accident and transferring risk exposure to third unprotected parties.

4. Lower speed limits are expected to impose long‐run time delays on all motorists; however, any higher speeds that increase the probability of an accident can impose substantial short‐run time delays when traffic is interrupted by a specific accident. On balance, the time costs are higher with an across‐the‐board reduction in maximum speed limits.

5. Another context is one where the maximum speed limits vary for the same class of road. Australia does not have such variation, which denies the opportunity to develop an econometric model to study explicitly the relationship between hours of travel, vehicle‐km travelled, fatalities and speed limits. Recent research by Ashenfelter and Greenstone (2004) has demonstrated empirically that without this extra information on variations in legal speed limits (which exists in the USA between states), it is not possible to identify a model in which there exists an explanatory variable that affects speeds and that only affects fatalities through its effect on speeds. Given the high degree of correlation in safety data, this is a serious issue, and one that cannot be handled appropriately in an Australian context using traditional methods of regressing accidents against speeds. This is one amongst a number of reasons why an alternative approach to establishing the economic externality costs of accidents is proposed herein.

6. Public policy in many countries adopts an accident‐flow elasticity of zero, which implies that no significant accident externality exists. This is questionable and research in economics in particular has argued since Vickrey’s (1968) pioneering research that this elasticity is likely to be in the range 0.25 (Newbery, 1988) to 1.5 (also Fridstrøm, 1999).

7. For further details, see Steimetz (2004).

8. As well as within an individual on different occasions.

9. For commercial‐in‐confidence reasons, the specific models used in the evaluation of specific toll roads in Sydney cannot be reported. However, heterogeneity around the mean was explained by a number of respondent perceptions of the relative advantage of tolled versus free routes (e.g. better access to the city, fewer traffic lights) and heteroskedasticity of the variance of random parameters was influenced by trip length (km).

10. The Weibull(b, c) is: w = b*(‐logU)∧(1/c).

11. The caveats associated with a direct comparison with Steimetz’s findings, as pointed out by a referee, are as follows. (1) estimates from Steimetz (2004) are based on data from a toll‐road project that features toll lanes adjacent to the project’s non‐tolled lanes. In other words, these toll lanes are essentially part of the same highway as the non‐tolled lanes. In contrast, the present study exploits data based on choices between a tolled highway and non‐tolled arterial routes. These arterials might include other features, such as traffic signals, that can generate different types of unobservable influences on motorists’ choices. The use of a mixed‐logit specification, together with a toll route quality bonus to account for these unobservable effects in the current study, may ameliorate these concerns. (2) Motorist demographics, such as income levels, may differ substantially between the motorists observed in the present study and those in Steimetz (2004). (3) In making international comparisons, note that the empirical framework in the present study and that in Steimetz (2004) are likely to capture the value of any non‐travel delay disutility associated with increased traffic densities. Though it is argued that the bulk of this disutility is attendant to accident risk and defensive effort, differences in cultures and attitudes toward traffic congestion may produce different ‘value of density’ estimates under identical density and toll conditions. The present author suggests that Sydney is not unlike the Southern California setting in this respect (the author having lived in both locations in the 1990s).

12. The comments of a referee were very useful in emphasizing this point.