Estimating the effectiveness of a vehicle miles travelled tax in reducing particulate matter emissions

Abstract

This study estimates the effectiveness of a vehicle miles travelled (VMT) tax in controlling mobile-source emissions of particulate matter (PM_2.5) in a non-attainment area located in northern Utah. Using a recently updated household-level dataset, the study finds no evidence of an endogenous relationship between choice of vehicle type and VMT. VMT elasticities are also estimated with respect to cost per mile that are in some cases larger in magnitude than those reported in previous studies. Based on vehicle emissions tests performed by the Houston Advanced Research Center, the study estimates the reduction in particulate emissions that would occur with two different sets of VMT tax rates. Principal findings are that a VMT tax rate of $0.003 per passenger car mile and $0.01 per light-duty truck mile (resulting in a mean annual tax burden of $128 per household in the first year) would reduce annual particulate emissions by between 7% and 11%, depending upon the degree of heterogeneity in household driving behaviour. Assuming constant elasticity, this means that at tax rates of $0.006 and $0.02 per mile for passenger cars and light-duty trucks, respectively (resulting in double the mean annual tax burden), annual particulate emissions would be reduced by between 12% and 23%. Both the advantages and limitations of the VMT tax are discussed.

Keywords:

• vehicle miles travelled (VMT)
• tax
• VMT elasticity
• particulate matter

Acknowledgements

The authors wish to thank Paul Jakus and John Keith for helpful comments on an earlier draft of the paper.

Notes

1. Ostro et al. (2006) find that an increase in PM_2.5 by 10 μg/m³ corresponds to a 0.6% increase in risk of mortality.

2. Conventional wisdom suggests that emissions from area sources (such as residential wood smoke and industrial production) are typically the largest primary contributions to PM_2.5 pollution. However, primary and secondary contributions (the latter in the form of nitrogen and sulfur oxide emissions) combine to make mobile sources the largest overall contributor to PM_2.5 pollution, particularly in densely populated areas (Malek et al., 2006, Utah Department of Environmental Quality 2008, US Environmental Protection Agency 2008). This is indeed the case for the study area here, Cache County, Utah, and at least partially explains why remediation programmes are focused on reducing emissions from mobile sources (Cache Chamber of Commerce 2008).

3. To the extent that particulate emissions are correlated (i.e. occur jointly with) other mobile-source criteria pollutants, e.g. carbon monoxide and ozone, etc. as well as with mobile-source global pollutants such as carbon dioxide, the tax policy examined in this paper will also work to control these emissions.

4. For example, see Walls et al. (1993), Sevigny (1998), and West (2004). In the fourth section of this study, elasticity estimates are compared with estimates from these studies.

5. However, VMT elasticity generally increases for light-duty trucks as the move is made from one- to three-or-more vehicle households.

6. The area, Cache County, recently obtained non-attainment status after several consecutive years of exceeding both the old and new maximum-allowable 24-hour standards. The county borders the state of Idaho. Approximately 100,000 people reside in the county, with the largest concentration (40%) in the city of Logan.

7. For reasons discussed at length in the fourth section, cars, vans and SUVs have been put together in the same grouping.

8. Although it could potentially be levied seasonally, a gas tax cannot be applied strictly to locally occurring VMT, even with statistical averaging. An additional disadvantage with a gas tax is the inability of the taxing authorities to prevent drivers from purchasing their gas in neighbouring counties (although Cache County is large enough such that this might not be problem in the larger metropolitan areas). However, a gas tax does provide control for pollution caused from idling engines, whereas a VMT tax does not.

9. The need to adjust the VMT tax for seasonality and location suggests that the efficient VMT tax is some proportion (either less than or greater than one) of the estimated annual VMT tax for Cache County households derived in the fifth section. Designing the infrastructure necessary to overcome the practical problems of administering a VMT tax (such as the potential roll-back of odometers, seasonality and ‘locationality’ mentioned above) is beyond the scope of this paper.

10. Fullerton and West (2000) consider heterogeneity with respect to vehicle-based functions (i.e. the shape of the emissions-per-mile and miles-per-gallon functions) and household behaviour (i.e. the correlation between VMT and vehicle attributes). The former type of heterogeneity is what is referred to as vehicle specific and the latter type is referred to as household specific. Since VMT varies by household, a VMT tax is by its very nature household specific.

11. In a related paper, Fullerton and West (2002) consider three alternative first-best tax scenarios (in comparison with an emissions tax): a vehicle-specific gas tax; a vehicle-specific tax based on attributes such as engine size, pollution control equipment (PCE), and VMT; and a three-part tax/subsidy on gas, engine size and PCE. They find that in order for a first-best gas tax to be feasible, the attributes of each vehicle would have to be identifiable at the pump.

12. Superscripts on functions identify household i or vehicle type j, while subscripts represent partial derivatives with respect to the identified variable. For expository purposes it is assumed that function f is deterministic. In reality, this function is subject to uncertainty with respect to technological and environmental factors. It is also assumed that the emissions mix uniformly in forming particulate concentrations in the atmosphere.

13. The 2001 RTECS data updates the dataset used by Sevigny (1998).

14. Various interaction terms, formed from combinations of these variables, are described when presented in the fourth section.

15. Full descriptions of this and all other variables used in the analysis of the fourth section are available upon request from the authors.

16. The regression models discussed in the fourth section were also estimated with the dummy variables car_j , j = 1,2,3, where car_j  = 1 if the vehicle is passenger car and 0 otherwise. Miles-per-gallon were also used to proxy for vehicle type. The results using these vehicle-type measures were qualitatively similar to those using truck_j .

17. See Bento and Goulder (2007) for a joint-estimation approach to obtaining elasticity estimates for two-vehicle households.

18. Nlogit version 3.0.10 was used to generate the following results.

19. A null hypothesis for a test of the endogeneity of truck ₁ can therefore be expressed as μ ₁ being a mean-zero, normally distributed error term.

20. Results are provided based on equation (3a) (as well as equation (5a) to follow), as the results for (3b) (and (5b) to follow) are (respectively) qualitatively similar. The results for (3b) and (5b) are available upon request from the authors.

21. The measure of commuting is admittedly qualitative rather than quantitative (i.e. cm_j merely indicates whether a household has stated that the vehicle is used primarily for commuting purposes, not the specific extent to which it is used as such). Therefore it is an imprecise measure of commuting. The authors thank an anonymous referee for pointing this out to them.

22. Note that intercm_j  = cm_j  · lavecost_j , interlinc_j  = lowinc · lavecost_j , interminc_j  = midinc · lavecost_j , and intert_j  = truck_j · lavecost_j , j = 1,2. The inclusion of lavecost ₂ in (5a) and lavecost ₁ in (5b) controls for cross-price effects.

23. In this case (4) can be written as

24. The statistical significance of the Wald Chi-Square statistic (at the 1% level) indicates that the SUR restrictions of homogeneity and Walrasian demand yield more efficient coefficient estimates.

25. The authors do not have a good theoretical explanation for why both vehicles behave as luxury goods for these particular households.

26. Thus the scale effect outweighs the substitution effect, as discussed in the first section. Note that this occurs solely for passenger cars (and light-duty trucks if they are the household's first vehicle), as the coefficient estimate for intert ₂ adds a ‘base’ of −0.91 to the non-commuter VMT elasticity for each income category.

27. (5b) was also estimated without the cm_j and truck_j variables included. In this case, the coefficient estimate for intercm_j became negative, suggesting that households which use their second vehicle for commuting purposes have higher VMT elasticities.

28. Recall that the data covers households in 2001, while the previous studies' results are based on household samples conducted in the mid-1990s and earlier.

29. West's (2004) dataset is compiled from the 1997 Consumer Expenditure Survey and the California Air Resources Board Surveillance Program.

30. Yu and Qiao (2004) do not report separate emissions results for vans and SUVs. Since light-duty trucks, vans, and SUVs generally match in terms of vehicle weight, engine size and fuel burn, vans and SUVs are put together with light-duty trucks rather than with passenger cars for this particular analysis.

31. A simpler approach would have been to multiply the product of Yu and Qiao's (2004) PM_2.5 emissions estimates and Utah DMV's (2006) estimate of average VMT for trucks and passenger cars by the total number of light-duty trucks and passenger cars, respectively, in the Cache County fleet. However, this approach seems less precise because only an average VMT is used rather than actual VMTs reported by households themselves in a survey. Ideally, the Utah Department of Environmental Quality would calculate an annual estimate of particulate emissions (and perhaps concentrations themselves) attributable to mobile sources in Cache County. To the authors' knowledge, such estimates do not currently exist.

32. These baseline tax rates are similar to those purposed by Sevigny (1998).

33. The heterogeneous household model for this analysis distinguishes commuter and non-commuter vehicles according to household income levels, i.e. a commuter vehicle from a high-income household, commuter vehicle from a middle-income household, commuter vehicle from a lower-income household, non-commuter vehicle from a high income household, etc.

34. In cases where a commuter-vehicle VMT elasticity estimate is positive, the estimate is set to zero to calculate the particulate emissions reductions presented in Table 7.

35. The two assumptions are possibly grandiose. The former depends on inter alia daily weather and traffic flow conditions. To the authors' knowledge, there are no formal studies addressing the link between emissions and concentration. The latter depends on daily driving habits of households. Understanding these habits would require more detailed survey information than is presently available.

36. VMT revenues might also be increased to obtain additional transportation dollars through the Federal-aid Highway Program administered by Department of Transportation's Federal Highway Administration (FHWA). The authors thank an anonymous referee for alerting them to this possibility.