Air Quality and Exercise-Related Health Benefits from Reduced Car Travel in the Midwestern United States

Background: Automobile exhaust contains precursors to ozone and fine particulate matter (PM ≤ 2.5 µm in aerodynamic diameter; PM2.5), posing health risks. Dependency on car commuting also reduces physical fitness opportunities. Objective: In this study we sought to quantify benefits from reducing automobile usage for short urban and suburban trips. Methods: We simulated census-tract level changes in hourly pollutant concentrations from the elimination of automobile round trips ≤ 8 km in 11 metropolitan areas in the upper midwestern United States using the Community Multiscale Air Quality (CMAQ) model. Next, we estimated annual changes in health outcomes and monetary costs expected from pollution changes using the U.S. Environmental Protection Agency Benefits Mapping Analysis Program (BenMAP). In addition, we used the World Health Organization Health Economic Assessment Tool (HEAT) to calculate benefits of increased physical activity if 50% of short trips were made by bicycle. Results: We estimate that, by eliminating these short automobile trips, annual average urban PM2.5 would decline by 0.1 µg/m3 and that summer ozone (O3) would increase slightly in cities but decline regionally, resulting in net health bene-fits of $4.94 billion/year [95% confidence interval (CI): $0.2 billion, $13.5 billion), with 25% of PM2.5 and most O3 bene-fits to populations outside metropolitan areas. Across the study region of approximately 31.3 million people and 37,000 total square miles, mortality would decline by approximately 1,295 deaths/year (95% CI: 912, 1,636) because of improved air quality and increased exercise. Making 50% of short trips by bicycle would yield savings of approximately $3.8 billion/year from avoided mortality and reduced health care costs (95% CI: $2.7 billion, $5.0 billion]. We estimate that the combined benefits of improved air quality and physical fitness would exceed $8 billion/year. Conclusion: Our findings suggest that significant health and economic benefits are possible if bicycling replaces short car trips. Less dependence on automobiles in urban areas would also improve health in downwind rural settings.


Research
The current fossil fuel-based transportation system of the United States negatively impacts human health by increasing air pollution and automobile accidents and by decreasing physi cal activity. Here, we consider how replacing short automobile trips with bicycle transport might yield health benefits through improved air quality and physical fitness, with a focus on the upper midwestern United States as our study region.
Both ozone (O 3 ) and fine particular matter ≤ 2.5 µm in aerodynamic diameter (PM 2.5 ) in the ambient air exacerbate bronchitis and asthma and may contribute to cardio vascular mortality (Brunekreef and Holgate 2002). Asthma affects 8.2% of U.S. citizens, and an estimated 10 million adults have diag nosed chronic obstructive pulmonary disease (COPD) (Centers for Disease Control and Prevention 2009). In addition, recent estimates attribute 63,000-88,000 pre mature deaths per year due to PM 2.5 [U.S. Environmental Protection Agency (EPA) 2010c]. In the United States, onroad vehicles are responsible for about 26% of volatile organic compounds (VOCs) and 35% of nitrogen oxide (NO x ) emissions (U.S. EPA 2005c(U.S. EPA , 2005d. NO x and VOCs combine to form O 3 and contribute to nitrate and secondary organic aerosols, impor tant components of PM 2.5 . Nearly 240 U.S. counties, with > 118 million total residents, exceeded U.S. EPA O 3 standards in 2011, and > 200 counties (> 88 million total residents) failed to meet PM 2.5 standards, in part because of pollution from short car trips (U.S. EPA 2011a, 2011b).
Transportrelated inactivity, that is, the use of motorized transport rather than walk ing and bicycling, has been linked to increased mortality and decreases in healthy life years, with the greatest impacts on chronic diseases including heart disease, stroke, colon cancer, diabetes mellitus type 2, obesity, breast cancer, and osteoporosis [World Health Organization (WHO) 2002]. Carlson et al. (2009) esti mated that 32.4% of the U.S. population is fully inactive (no moderateintensity or vigorousintensity physical activity lasting at least 10 min at a time), while only 33.5% is physically active, defined as 30 min/day with moderateintensity activity, ≥ 5 days/week. In a recent Dutch study, Johan de Hartog et al. (2010) concluded that shifting from short car trips to bicycle trips would reduce allcause mortality, with estimated reductions in mortality due to increased physical activity that were nine times greater than estimated increases in mortality due to increased pol lution inhalation and trafficrelated fatality estimates in the Netherlands.
In the United States, 28% of all car trips are ≤ 1.6 km (1 mi), which is the dis tance that a typical European would walk (European Commission 2001;Pucher and Dijkstra 2003). Another 41% of all trips are ≤ 3.2 km (2 mi), a distance that many Europeans would be as likely to bicycle as to walk (European Commission 2001;Pucher and Dijkstra 2003). If we use European travel behavior as a point of comparison for walking and bicycling activity for the United States, these data suggest that many car trips in the United States could be avoided.
Amplifying the potential benefits of increased bicycle use is the nonlinear relation ship of vehicle emissions to travel time. A large fraction of emissions (25% of VOC and 19% of primary PM 2.5 ) are emitted in just the first few minutes of automobile operation, often known as "cold start," before pollution control devices operate [Federal Highway Administration (FHWA) 2006]. Because emissions control systems reach operating tem perature only after several miles of travel and typically cool below operating range in < 1 hr (Singer et al. 1999), reducing the number of short trips could disproportionately curtail pollutant emissions from onroad vehicles.
In the present study, we quantified the potential health and monetary bene fits of replacing short (≤ 4 km one way) car trips with travel by bicycle (50% of trips) in the 11 largest mid western metropolitan sta tistical areas ( Stone et al. (2007). We estimated changes in regional emissions and air quality, as well as resulting health benefits, across the upper mid western states [see Supplemental Material, Figure 1 (http://dx.doi.org/10.1289/ ehp.1103440) for a map of the area]. In addi tion, we estimated the benefits of increased physical activity using the Health Economic Assessment Tool (HEAT) for cycling devel oped by the WHO (Rutter et al. 2007).

Methods
We estimated that eliminating short car trips (≤ 8 km round trip) in urban areas of Illinois, Indiana, Michigan, Minnesota, Ohio, and Wisconsin would reduce residential vehicle use by 20%. This estimate is based on a census tract level travel and mobile emissions inven tory by Stone et al. (2007), who combined 1995 Nationwide Personal Transportation Survey (NPTS) responses (FHWA 1997), demo graphic modeling of household vehicle travel, and the U.S. EPA MOBILE6 emis sions factor model (U.S. EPA 2004b). From that contemporary emissions inventory, we estimated current emissions levels if all round trips of ≤ 8 km in urban and suburban cen sus tracts were made using alternate modes of transportation. To inform the potential impact of a range of realistic policies and choices, we used these estimated reductions to quantify the maxi mum potential response to a change in travel behavior. Although arbitrary, this assumed reduction in short auto trips would be consistent with the use of active (cycling or walking) transportation in European cit ies similar in density and population to the MSAs considered here. These values represent theoretical upper bounds on shorttrip trans portation choices under current travel patterns and population density. We assume that no change occurs in rural travel because distances between residential and commercial areas are typically too great for bicycling or walking and because rural populations are too low to sup port mass transportation.
Specifically, we compared transporta tion modes used in the studyarea cities, with populations ranging from 837 persons/km 2 in Grand Rapids to 4,884 persons/km 2 in Chicago (average 2,051 persons/km 2 ), to five European cities with similar population densi ties (range 901-5,971, average 2,910 persons/ km 2 ) [see Supplemental Material, Table 1 (http://dx.doi.org/10.1289/ehp.1103440)]. Although public transportation use was simi lar, only 39% of trips were made by auto mobile in the European cities, compared with 80% of trips in the Great Lakes region. Although the configurations and historical growth patterns of the European cities differ from their American counter parts, the fact that half of all trips used active transportation suggests that active transport for 50% of short trips is feasible for similar travel distances in mid sized American cities of similar density and that greater active transportation need not be limited to areas of highest density.
We estimated changes in emissions only for onroad lightduty passenger vehicles with internal combustion engines and only for round trips ≤ 8 km. We modeled changes in primary emissions (including NO x , carbon monoxide, sulfur dioxide, ammonia, VOCs, elemental carbon, organic carbon, and pri mary fine and coarse particulate matter) from all stages of vehicle operation, as well as emis sions from evaporation, brake dust, resus pended road dust, and refueling. Reducing the number of short trips further lessens the frequency of cold starts from 59.9% to 21.9% of trips in urban tracts and from 55.6% to 20.3% in suburban tracts, with correspond ing reductions in VOC and NO x emissions. We mapped emissions from the censustract level to the 12 × 12 km 2 model grid by area weighted averaging using the U. We estimated changes in ambient air PM 2.5 and O 3 concentrations using hourly regional chemical transport simulations with the Community Multiscale Air Quality Model (CMAQ), version 4.6 (Byun and Schere 2006), driven by meteorology from the weather research and forecasting model for the full year of 2002 . Simulations with CMAQ were conducted on a 12 × 12 km 2 grid and included gas phase, aqueous, and heterogeneous chemical reactions and equilibrium aerosol thermo dynamics. We fol lowed the model configuration used by Spak and Holloway (2009), with boundary condi tions from a 36 × 36 km 2 simulation over continental North America.
We used the Environmental Benefits Mapping and Analysis Program (BenMAP) version 4.0.35 (U.S. EPA 2010a) to estimate health impacts due to CMAQsimulated changes in ambient air pollution resulting from reduced car travel. Because BenMAP addresses both mobile and stationary sources (U.S. EPA 2004a(U.S. EPA , 2008, it has been used to support the creation of environ mental regula tions in several countries. After air quality data is loaded into BenMAP, the program determines the change in ambient air pollution. BenMAP then uses concentration-response functions (CR) to calculate the relationship between the pollu tion and certain health effects, applying the relationship to the exposed population (Abt Associates 2010). Finally, BenMAP uses a "damage function" to estimate health costs and benefits from changes in air quality. A damage function quantifies the health benefits and economic value of reduced exposure to pollutants (Davidson et al. 2007).
BenMAP 4.0 (i.e., version 4.0.35) incor porates hourly air pollution data and county level baseline incidence rates for the following health outcomes: overall mortality, asthma exacerbations, chronic bronchitis, hospital admissions, acute myo cardial infarctions, acute and chronic respiratory infections, upper and lower respiratory infections, workloss days, and schoolloss days. Spatial speci ficity in base line incidence data varies by health outcome and location; where countylevel data are not available, BenMAP distributes state estimates to the county level using agespecific rates for each health outcome within each county. For mortality estimates, BenMAP combines nationallevel census mortality rate projec tions and countylevel agespecific incidence rates from 2006 with projected changes in study area populations to derive county level mortality rate projections for 2010. For the present study, BenMAP used statelevel hospitali za tion data to estimate countylevel incidence for Minneapolis/St. Paul, Chicago, and Indianapolis; countylevel incidence data volume 120 | number 1 | January 2012 • Environmental Health Perspectives for all cities in Ohio; and city hospital dis charge data for Milwaukee, Madison, Detroit, and Grand Rapids. For emergency room (ER) admissions, we used midwest regional incidence data for Detroit, Grand Rapids, Chicago, and Indianapolis; statelevel data for Minneapolis/St. Paul; countylevel data for Ohio cities; and hospital dischargelevel data for Milwaukee and Madison. For all cities, non fatal acute myo cardial infarction incidence rates were based on regional hospitalization data. All other health end point data were based on national figures.
BenMAP assigns monetary values to the reduction of adverse health effects based on national averages that do not reflect intra city or inter city variability in costs. The BenMAP analysis was conducted on the 12 × 12 km 2 grid, using 2010 census projection allocation to the grid by the U.S. EPA. Valuation is in 2010 dollars.
We combined air quality estimates for 2002 from CMAQ with 2002 U.S. EPA monitoring using spatial scaling by Voronoi nearest neighbor averaging (e.g., Chen et al. 2004). This pairing yields air quality inputs to BenMAP including complete spatial and temporal coverage by highresolution hourly modeling, constrained to match concentra tions observed near monitors. We then used the expertderived PM 2.5 CR functions, valua tion estimates, and pooling methods used for the U. Because multiple studies exist for each given health incidence, pooling techniques are often used to statistically combine the results. Using BenMAP, we ran each CR function and pool ing of incidence and valuation for each health end point in a 5,000member Monte Carlo ensemble. Sources of CR functions used in this analysis are presented in Tables 1 and 2. As standard practice, the U.S. EPA does not pool mortality studies. Thus, we used the Harvard Six Cities study (Pope et al. 2002) as BenMAP input for PM 2.5 mortality; that study included the most representative sites. We selected the 2010 population database to use in BenMAP because the sensitivity studies we conducted indicated that choice of year has no substantial impact (1-2% difference) on incidence of health threats. Values for incidence represent estimated incidence per adverse health effect avoided due to a change in air pollution in the given city per year; savings are presented in millions of dollars. Values in parentheses are 95% confidence intervals, and all changes are annualized. a Change in PM 2.5 (µg/m 3 ) was calculated as area averaged and reported with a range of minimum and maximum grid cell values; data for PM 2.5 -related health effects estimated in this analysis (and the source of the PM concentration-response functions used to estimate the change in incidence) are from Abbey et al. (1995), Dockery et al. (1996), Ito (2003), Laden et al. (2006), Moolgavkar (2000aMoolgavkar ( , 2003, Norris et al (1999), Ostro (1987), Ostro and Rothschild (1989), Ostro et al. (2001), Peters et al. (2001), Pope et al. (1991Pope et al. ( , 2002, Schwartz and Neas (2000), Sheppard (2003), and Vedal et al. (1998). b Respiratory problems include upper and lower respiratory symptoms, hospital admissions (respiratory), emergency room visits (respiratory), and cases of acute bronchitis. c Cardiovascular problems include non fatal acute myocardial infarctions and cardiovascular hospitalizations. d Yearly work-loss-day incidence based on estimates from the 1996 National Health Interview Survey (Adams et al. 1999).
To address the potential health and eco nomic cobenefits that would result if half of all short trips were made by bicycle, we used HEAT. This model uses relative risk data (Anderson et al. 2000) to estimate cost savings from reduced allcause mortality. Controlling for socio economic variables (e.g., age, sex, smoking) and leisure time activity, HEAT calculates risk reduction for days spent cycling based on estimates of total number of days cycled, distance, and average speed (Rutter et al. 2007).
We used HEAT analysis to estimate the monetized health benefits associated with the conversion of onehalf of short trips (< 8 km round trip) by car to be made by bicycle. This represents 10% of vehicle miles traveled (VMT) for the region. We used the U.S. EPA value of a statistical life ($7.4 million) (U.S. EPA 2010b) and the annual percentage of allcause work ingage mortality [0.00390; 95% confidence interval (CI): 0.00277, 0.00503] (Wilkinson and Pickett 2008). We assumed an average of 124 days of cycling per year, HEAT's default value (Rutter et al. 2007), which is representa tive of the climate of the upper midwest, where bicycle commuting is most common from April through October. We also assumed that only 50% of these trips would be under taken by peo ple who do not currently cycle, thus excluding the small percentage of the population already benefitting from cycling, as well as elderly indi viduals or those physically unable to bicycle. We used the NPTS average commute distance for each MSA (from 3.34 to 3.98 km) with an average speed for commuter cyclists of 14 kph. Finally, we used the HEATrecommended default percentage (90%) of cyclists completing a round trip each day.

Results
Simulations yielded unique hourly esti mates of surfacelevel PM 2.5 throughout the year ( Figure 1A) and O 3 during the warm season (1 May 30-September) ( Figure 1C) on a 12 × 12 km 2 grid for 2002. The CMAQ simulations described here captured spa tial and temporal variability in PM 2.5 [see Supplemental Material,  Table 3) when compared with U.S. EPA monitoring data throughout the region, with performance for PM 2.5 and O 3 both exceeding community and U.S. EPA expectations for chemical transport modeling in policy and research applications.
We estimated that substitution of non emitting modes for short trips would achieve average annual reductions in the 24hr aver age PM 2.5 concentrations considered in U.S. PM 2.5 regulations ( Figure 1B). Regional O 3 would also be reduced throughout the May-September summer season (calculated based on daily maximum 8hr and 1hr averages, consistent with U.S. O 3 regulations) but day time O 3 would increase in the largest cities Values for incidence represent estimated incidence per adverse health effect avoided due to a change in air pollution in the given city per year; costs are expressed as negative and benefits as positive (millions of dollars). Values in parentheses for incidence and savings are 95% confidence intervals (in most cases rounded to nearest decimal), and all changes are annualized. volume 120 | number 1 | January 2012 • Environmental Health Perspectives because of VOClimited O 3 production con ditions in urban environments ( Figure 1D). Effects of transportation on O 3 concentra tions within the MSAs are complex because of the non linear inter play of emissions and meteorology in atmospheric chemistry and transport, whereby local ambient O 3 concen trations often increase in response to reduc tions in NO x and/or VOC emissions (Sillman 1995). In our emissions inventory, motor vehicles were responsible for most of the NO x (70-98%) and VOC (40-95%) emissions in the MSAs, with the highest percentages of emissions from motor vehicles in the most urbanized areas. Although Figure 1B and D show longterm averages (annual for PM 2.5 and summer for O 3 ), we used hourly values from CMAQ to estimate the potential health benefits of increased active transport.

Fine particulates (PM 2.5 ).
We observed changes in PM 2.5 and O 3 , associated health outcomes, and monetary savings for each MSA and for the combined total of all grid cells out side the 11 MSAs (Tables 1 and 2). We esti mated that eliminating short car trips would reduce annual average PM 2.5 across the study region by 0.08-0.15 µg/m 3 (1.0-2.0%) in most MSA urban centers. In the upwind MSAs of Madison and Minneapolis/St. Paul, which would see little bene fit from PM 2.5 reductions in other cities, we estimated that PM 2.5 would be reduced by 0.05 µg/m 3 ( Figure 1B). Nearly all of the estimated reduction in PM 2.5 would be due to decreases in secondary aerosols, espe cially nitrate formed from NO x and secondary organic aerosols from VOCs. Primary particle emissions from motor vehicles are negligible, so the reduced VMT scenario would not signifi cantly affect this smaller fraction of PM 2.5 mass. Reductions in PM 2.5 in urban areas and down wind would be greatest during highpollution episodes exceeding the 24hr average PM 2.5 NAAQS. In urban grid cells, the average esti mated reduction during NAAQS exceedances was 0.20 µg/m 3 , equivalent to the maximum change in annual average PM 2.5 in Chicago [see Supplemental Table 4 (http://dx.doi. org/10.1289/ehp.1103440)]. In addition, we estimated that the reduction in short auto trips would result in one fewer exceedance per year in a typical urban grid cell and a 5-25% reduc tion in the number of annual exceedances.
Our results indicate that adverse health outcomes related to PM 2.5 would be reduced in all MSAs (Table 1). Reductions in PM 2.5 related mortality across the midwest are shown in Figure 2A, with the total impact across the 37,000mi 2 region being 525 fewer deaths. We estimated that asthma exacerba tions would decrease annually by > 2,500 cases. In addition, there would be approximately 100 fewer COPD cases, whereas net respira tory symptoms, hospital admissions, and ER  visits would decrease by 94,186 cases annually. Regarding cardio vascular disease, there would be approxi mately 860 fewer cases of non fatal acute myocardial infarction and hospital admis sions. Savings from reduced annual mortality would reach almost $4.14 billion. Savings of > $7.5 million would result from fewer respi ratory cases, hospital admissions, and ER vis its, whereas a reduction in COPD would save > $39 million per year; reductions in non fatal acute myo cardial infarctions and cardio vascular hospitalizations would save > $54 million. We estimate that total savings from reducing adverse health effects due to PM 2.5 would be about $4.25 billion/year (95% CI: $598 mil lion-$11.2 billion). Projections suggest that PM 2.5 exposure would also be reduced in popu lations outside MSAs and that resulting reduc tions in adverse health effects would account for roughly 25% of the total benefit. Ozone. Estimated effects of eliminating short car trips on O 3 pollution vary in rela tion to the size and density of urban areas. For large urban areas, estimated daily 8hr maximum, 1hr maximum, and daily average O 3 concentrations during the May-September O 3 season generally increased in city centers, whereas concentrations decreased in suburbs, some smaller urban areas, and in areas down wind of the MSAs ( Figure 1D). Simulated changes in transportation and reductions in coldstart frequency would decrease total NO x emissions by 5-12% and total VOC emissions by 10-25%.
Although we estimate that NO x and VOCs would both be reduced, the response to NO x reductions would be more pronounced, result ing in increased O 3 in urban cores, consistent with previous studies in the region (Sillman 1995). Changes in estimated O 3 concentra tions were greater during the warmest months (July-August) when concentrations are highest, with increases and decreases of up to 2 ppb. We estimate that daily 8hr maximum O 3 would increase on a populationweighted basis (Table 2) but that areaaveraged O 3 levels would decrease in every MSA. BenMAP analy sis indicated net regional savings from declines in mortality, schoolloss days, hospitalizations, ER visits, and acute respiratory symptoms, but some increases in costs in cities such as Chicago, Cleveland, Columbus, Milwaukee, and Minneapolis/St. Paul due to changes in O 3 levels. Costs resulting from O 3 increases due to reduced VMT were statistically sig nificant for only Chicago and Minneapolis/St. Paul, but estimated savings from PM 2.5 reduc tions were greater than increased costs due to O 3 in all cities.
We estimated that areas outside the MSAs would experience net benefits for all O 3 related health outcomes. For nine of the cities (exclud ing Chicago and Minneapolis/St. Paul), we estimated a potential reduction of approxi mately 30,000 cases in acute respiratory symp toms associated with the potential changes in O 3 (resulting in savings of almost $1.9 million) and 8,632 fewer schoolloss days (savings of almost $822,000). This distinct reduction in acute respiratory symptoms to areas outside the MSAs is shown in Figure 2B.
Estimated changes in health outcomes due to changes in O 3 are less correlated with MSA density or size than estimated changes due to reduced PM 2.5 , particularly for out comes related to daily peak values, such as acute respiratory symptoms. Instead, esti mated changes in O 3 related health impacts were often more pronounced in smaller MSAs such as Dayton and Grand Rapids, reflecting differences in total VOC:NO x ratios and the degree to which reductions in local motor vehicle emissions would alter them. Thus, estimated effects of eliminating short car trips on population O 3 exposures are highly sensi tive to urban size, density, and travel patterns.
Benefit from physical activity. Based on WHO HEAT analysis, we estimated that com pleting 50% of short trips by bicycle would result in average annual savings of > $2.5 bil lion for short suburban bicycle trips and nearly $1.25 billion for short urban trips (Table 3), for a total of approximately $3.8 billion in bene fits across an estimated population of 2 million people and a reduction in pre mature mortality of almost 700 deaths/year.

Discussion
In the study region with a population of 31.3 million, we estimated that eliminating short car trips and completing 50% of them by bicycle would result in mortality declines of approximately 1,295 deaths/year (95% CI: 912, 1,636), including 608 fewer deaths due to improved air quality and 687 fewer deaths due to increased physical activity. Changes in PM 2.5 and O 3 would result in net health bene fits of $4.94 billion/year (95% CI: $0.2 billion, $13.5 billion). Completing 50% of short trips by bicycle would yield $3.8 billion/year in savings (95% CI: $2.7, $5.0 billion), about $1.5 billion less in savings than from reductions in air pol lution. We estimate that the combined benefit from improved air quality and physical fitness for the region would exceed $8.7 billion/year, which is equivalent to about 2.5% of the total cost of health care for the five midwestern states in the present study in 2004 (Kaiser Family Foundation 2004).
Of course, an added benefit of remov ing 20% of VMT from the region is also reduced emissions of greenhouse gases that cause global climate change. The annual reduction would be > 1.8 teragrams carbon dioxide (CO 2 ) (3.9 billion pounds), using the fleet average passenger car fuel economy Few studies have addressed how changes in behavior can affect air quality (Frank and Engelke 2005;Frank et al. 2000), and none have quantified the potential benefits of travel behavior change for pollution control. Comparison with prior BenMAP cost-benefit regulatory analyses suggests that health bene fits from reduced air pollution through behav ioral changes in personal transportation would be comparable with effects of such topdown meas ures as the Clean Air Interstate Rule and the Nonroad Diesel Rule, both air quality regu lations having potential for substantial impacts on human health (Hubbell et al. 2009). The magnitude of regional impacts from urban travel mode substitution would be comparable with the annualized benefit of reducing O 3 nationwide to full compliance with the current 75 ppb NAAQS (U.S. EPA 2008).
Compliance with federal air quality stan dards through conventional measures such as emissions controls entails direct costs to govern ments and private industry. In contrast, chang ing personal travel behavior distributes costs and bene fits-both financial and otherwise-in a more complex manner, including potentially large personal savings for individuals given the high cost of vehicle owner ship and opera tion. However, in addition to public outreach, education, and incentive programs, drastic decreases in residential VMT would require infrastructure investments to support pedes trian and bicycle traffic, as well as increased public transit. For example, cities would need to designate bicycle lanes on streets, add bicy cle lanes or mixeduse nonmotorized paths, and provide additional signage, physical barri ers, bicycle traffic signals, and bicycle parking. Infrastructure costs for converting existing road ways to bicycle lanes in the United States range from $2,500 to $50,000/block, depending on the infrastructure needs. In 2010 Portland, Oregon, converted 10 blocks of hightraffic streets to include twoway bike lanes at a cost of $10,000/block, reducing motor vehicle traf fic by one lane. In 2011, Chicago added pro tected bicycle lanes with flexible marker posts and a parking lane for automobiles along four blocks, including a bridge, at a cost of $140,000 (City of Chicago 2011). Increasing this cost estimate to $100,000/block, double the U.S. average cost per mile for bike lane conversion and addition, the $2 billion in health cost sav ings in the MSA of Chicago alone could ret rofit 20,000 blocks (2,500 mi or 4,020 km) with bike lanes. The greater Chicago metro politan area has > 23,500 mi of urban roads, not including interstate or freeways (Illinois Department of Transportation 2009), so the health care savings could cover the costs of add ing bike lanes to every road in 1-10 years.
Although U.S. pedestrians and cyclists may be at higher risk of mortality than their Dutch counterparts (Pucher and Dijkstra 2003), the Dutch results provide a model for safer walking and cycling. Seven of the cities studied here-Chicago, Columbus, Dayton, Indianapolis, Madison, Milwaukee, and Minneapolis/St. Paul-have earned bicycle friendly rankings from the League of American Bicyclists because they actively support bicy cling by providing safe accommodation for cycling and encouraging people to bike (League of American Bicyclists 2010). Thus, some U.S. communities may be more likely than others to exhibit charac teristics of Dutch cities that make bicycling feasible. There is already an observed trend of increasing bicycle share across all of the 11 midwestern MSAs, one that is consistent and very large (U.S. Census Bureau 2009). Moreover, there is evi dence that U.S. cities with enhanced levels of active transport experience health benefits. Pucher et al. (2010) found that cities with the highest rates of commuting by bicycle or on foot have obesity and diabetes rates 20% and 23% lower, respectively, than cities with the lowest rates of active commuting.
Strengths, limitations, and uncertainties. Our research, for the first time, has joined models of health effects (BenMAP), census based vehicle use and emissions (PLUTO), and regional air pollution (CMAQ) to link highly localized changes in travel behavior to regional health outcomes. We also used the newest version of U.S. EPA BenMAP (4.0), which includes baseline incidence rates at the county (versus the regional) level, thus providing greater local specificity than previously possible. Our results may be a conservative estimate of pollution reductions. We did not evaluate changes in exposure for people who live or work near highways, nor did we assess health effects from decreases in other pollutants (e.g., carbon monoxide, sulfur dioxide) or the synergistic effects of combined changes in O 3 and PM 2.5 . We would expect the reduction in the num ber of automobiles on the road at any given time to change average speeds and resultant emissions, with variable effects on arterial and local roadways. Comprehensive analysis would require traveldemand modeling (e.g., Bowman and BenAkiva 2001) incorporating traveler decision making, spatially specific changes in roadway and transit networks, demographic information, and employment data to calcu late those differences in vehicle activity. Finally, health impacts from changes in longrange transport of O 3 and PM 2.5 to states downwind of the modeling domain and to neighboring Canadian regions were not analyzed.
Our health benefits analysis also may be conservative because, following current U.S. EPA practice, we used total PM 2.5 mass and did not differentiate between aerosol spe cies. Recent epidemiological studies suggest that trafficrelated emissions may contain more hazardous particulate chemical components. Gent et al. (2009) found more frequent asthma symptoms and inhaler use in children after exposure to PM 2.5 emissions attributable to motor vehicles compared with emissions from other sources. Bell et al. (2009) found differing associations between cardio vascular and respi ratory hospitalization across various chemical species of PM 2.5 . Particles comprising vana dium, nickel, and elemental carbon showed the strongest associations (vanadium and nickel come primarily from transportation emissions). However, because these epidemiological studies included high diesel truck traffic and its specific emissions profile, these results have slightly less bearing on our analysis of decreases in light duty automobile emissions.
Our estimates for physical fitness bene fits stemming from bicycling 50% of short car trips (≤ 8 km) may under estimate the full benefits of removing these car trips. Not included are the remaining trips that presumably would be achieved by some form of mass transportation or direct walking for very short trips. According to the 2001 National Household Travel Survey, Americans who use mass transit spend a median of 19 min daily walking to and from transit (Besser and Dannenberg 2005). Accounting for fitness benefits from this mode of active transport would involve complex geo spatial modeling. Future analyses should consider geo graphic information system (GIS) technolo gies in conjunction with energy expenditure measure ment tools, such as accelerometers or biometric monitors, to more accurately assess the speed, distance, intensity, and terrain of the cyclist (Bonnel et al. 2009). Finally, for urban planning purposes, assumptions for determin ing levels of benefits for new bicyclists will stem from cityspecific estimates of current bicycling levels and city wide demographics. We used current European bicycling levels to guide our maxi mum benefit level potentially achievable.
In our study we used chemical transport modeling simulations and empirical CR func tions, an experimental framework that adds incremental uncertainty at each step: in the emissions inventory, modeled meteorology, and processes included in the chemical trans port model. In addition, the ability of the model to reproduce observed ambient surface level O 3 and PM 2.5 and their respective sensi tivities to emissions changes adds uncertainty. We used the same suite of response functions and pooling chosen by the U.S. EPA for air pollution rule making; however, the empirical epidemiological CR functions of BenMAP and the choice of valuation estimates are an addi tional source of uncertainty. The valuation esti mates are a function of BenMAP, based on the configuration used by the U.S. EPA. Sensitivity analysis by the California Air Resources Board confirmed that the mean and distribution of premature mortalities from longterm exposure to PM 2.5 are not sensitive to the randomeffects pooling of CR functions (Tran et al. 2009). We found few outliers among the individual CR calculations that contribute to the reported pooled values. Although we chose to simulate a year (2002) that is representative of the regional climate of the past decade, the magnitude of bene fits achieved in any given year depends on inter annual variability in meteorology and the resultant ambient air quality.

Conclusion
Our study demonstrates that reduced car travel and enhanced bicycle commuting in urban areas can improve health outcomes within urban, suburban, and even in downwind rural areas. Our results demonstrate that reduced car travel can benefit air quality, human health, and the economy.

correction
In the manuscript originally published online, the weight of the annual reduction of CO 2 was noted in the "Discussion" as "3.9 trillion pounds" instead of "3.9 billion pounds," and information on health bene fits accruing outside the MSA regions was inadvertently omitted. Information for out side the MSA regions and for subsequent savings for the entire region is now included in Tables 1 and 2, and all values have been corrected here.