Health benefits of decreases in on-road transportation emissions in the United States from 2008 to 2017

Significance Despite decades of reductions in vehicle emissions in the United States, their impacts remain large, and the offsetting effects of different factors on benefits achieved in recent years are not well understood. We assess benefits from 2008 to 2017 on a fine spatial resolution using the latest epidemiological evidence and emissions inventories. We find that regulation continues to yield large benefits: $270 (190 to 480) billion in 2017 from reduced PM2.5-attributable mortality and greenhouse gas emissions. Traffic-related PM2.5-attributable mortality would have been 2.4 times as large in 2017 if vehicles had still been emitting at 2008 levels, accounting for most benefits. Urban passenger light-duty vehicles have become increasingly important, and major health gains require more stringent policies to curb their emissions.


Vehicle classifications
We use EPA's classification of vehicle types [1], shown in Table S1.
Vehicle types in NEI 2008 are different than in the other NEIs due to changes EPA's modeling process in 2011 (version 2) -when it moved from MOBILE to MOVES -and there is not a one-toone correspondence [1][2][3][4]. We converted vehicle types in the 2008 NEI to the types used in later versions according to EPA's conversion rule based on overall VMT in 2011 [1, pp. 124-126]. The difference seems small for LDVs, as it is limited to commercial light duty trucks, which represent only a small portion of LDV VMT. For HDVs, on the other hand, the changes in classification were more substantial.
We also had a very small number of counties where the data was inconsistent for one or more vehicle types, with either (i) emissions > 0 and VMT = 0 (or missing); or (ii) emissions = 0 (or missing) and VMT > 0. These are only a minor part of emissions. Emissions in case (i) summed up to just 2*10 -6 % of the total emissions across pollutants and years. Counties in case (ii) not exceeding 1.5% of either 2017 VMT or total emissions for any combination of pollutant, year, and vehicle type, with the exception of GHGs in California. In such cases, for each NEI year, we applied mean emission factors, with the exception of GHGs in California. If the county was outside of California, we applied U.S. means (excluding California). If the county was in California, we applied California's statewide mean emission factors. We separated California because the EPA does not model on-road emissions for the state of California -emissions are provided directly by the state [5]. GHG (i.e. CO2, CH4, and N2O) emissions in 2008 and N2O emissions in 2014 were missing for the state of California in the respective NEIs [3,6], so we complement them with data from the California Air Resources Board (CARB) [7]. GHG impacts are not location dependent, so in each case we applied state-wide emission factors for each of the four vehicle classes (motorcycles, LDVs, buses, and HDTs) using CARB's sector level 4 classification (Table S2). Although we do not adjust CARBreported emissions of GHGs in 2008 (or N2O in 2014), we note that CARB's CO2-eq. emissions in subsequent years (2011 and 2014) were lower than those reported in the respective NEIs by 13% and 11%, respectively.

Fleet composition and VMT adjustments
EPA's data indicates a substantial decrease in light commercial truck VMT over the period, while some types of heavy-duty trucks show a substantial increase (Table S3). This could be due to differences in vehicle classifications, even though the difference occurs between NEIs 2011 and 2014, and NEI 2011 v2 (which we use) already uses the current vehicle classification. Commercial light-duty trucks make up only a small percentage of the LDV VMT, but would make up a large percentage of the HDV VMT if they were heavy-duty vehicles.
We investigated the possible effect of vehicle classification in our results by creating an alternative VMT adjustment. Impacts for the 2017 EFs scenario are not affected since they are the impacts of emissions in 2017, but impacts in counterfactual scenarios for emission factors in previous years depends on adjusting VMT to 2017 levels. Our base results apply county-level vehicle-type emission factors per mile (EFs) to 2017 VMT from [8] -essentially scaling results from previous years according to changes in VMT of each vehicle type in each county. We construct an alternative adjustment, where we scale changes in VMT for each of: motorcycles, passenger cars, and all other vehicles combined. Classification of motorcycles and cars is unlikely to have changed, and we combine all other vehicle types together in case a portion of light-duty commercial trucks were re-classified as passenger trucks or heavy-duty vehicles.
In our base case, we estimate 48,200 deaths attributable to PM2.5 in 2017 in the 2008 EFs scenario, whereas the alternative VMT adjustment leads to 46,500, regardless of whether this alternative adjustment is done individually for each county's fleet composition or whether the same proportional adjustment is carried out based on the entire country's fleet composition change. VMT effects led to an increase in 22% since 2008 in our base case, a figure that drops to 17% in the alternative adjustment when the alternative adjustment is done at a county level, or to 18% when the alternative adjustment is done at a national level. This might seem counterintuitive since total fleet VMT increased by just 7% in the period, but a decrease of car VMT and an increase of larger vehicles leads to higher effects, even if it there is some uncertainty about whether these other vehicles are light or heavy-duty trucks. VMT for passenger cars -responsible for about a quarter of impacts in 2008 EFs prior to VMT adjustment -decreased by 15%, whereas the VMT for all other vehicles (except motorcycles) increased by 30%, and they make up about three quarters of impacts. This leads to VMT effects that are much larger than overall increases in VMT.

Supplementary Results
Figures S1-S11 and Table S4 show our supplementary results.