Neighborhood Social Inequalities in Road Traffic Injuries: The Influence of Traffic Volume and Road Design

Abstract

Objectives. We examined the extent to which differential traffic volume and road geometry can explain social inequalities in pedestrian, cyclist, and motor vehicle occupant injuries across wealthy and poor urban areas.

Methods. We performed a multilevel observational study of all road users injured over 5 years (n = 19 568) at intersections (n = 17 498) in a large urban area (Island of Montreal, Canada). We considered intersection-level (traffic estimates, major roads, number of legs) and area-level (population density, commuting travel modes, household income) characteristics in multilevel Poisson regressions that nested intersections in 506 census tracts.

Results. There were significantly more injured pedestrians, cyclists, and motor vehicle occupants at intersections in the poorest than in the richest areas. Controlling for traffic volume, intersection geometry, and pedestrian and cyclist volumes greatly attenuated the event rate ratios between intersections in the poorest and richest areas for injured pedestrians (−70%), cyclists (−44%), and motor vehicle occupants (−44%).

Conclusions. Roadway environment can explain a substantial portion of the excess rate of road traffic injuries in the poorest urban areas.

Injuries resulting from road crashes are leading causes of death and disability worldwide.¹ Although the number and rate of road deaths have decreased in industrialized countries, they remain a major public health burden, with approximately 40 000 and 3000 road deaths annually in the United States and Canada, respectively, and thousands more injuries.^1,2 For pedestrians, decreases may reflect a reduction of the exposed population, as, currently, fewer people walk as a mode of transportation.^3–5

There are significant social inequalities in road crashes, injuries, and deaths between and within countries.^1,6–8 Within countries and cities, motor vehicle injury and death rates have been shown to vary according to individual and neighborhood socioeconomic positions, with greater rates among the least well off.^9–13

Although many different factors, related either to individuals, vehicles, or the environment, contribute to creating such social inequalities, they should be understood in light of some mechanisms involved in the occurrence of road traffic injuries (RTIs). First, moving vehicles are the primary cause of road crashes: deaths and injuries result from the transfer of a motor vehicle's kinetic energy at a rate that exceeds the human body's protective capacity.¹⁴ Second, the burden of RTIs on population health is related to exposure to risk of crash.¹ Risk exposure can be estimated by distance traveled for drivers or traffic volume for streets and intersections.^15,16 Two California studies on neighborhood exposure to motor vehicles showed a greater likelihood of higher traffic volumes in the poorest census block groups and around schools in deprived areas.^17,18 Third, the number of injured pedestrians and cyclists is also related to the number of people exposed.^19,20 Thus, in a given environment, the more people walking, the more injured pedestrians. Fourth, the physical environment has a strong influence on the likelihood of injuries.²¹ Road widening increases crashes, whereas traffic calming and 20 mile per hour zones greatly reduce their occurrence.^22–26 In London, United Kingdom, the deprived areas have a larger proportion of traffic-calmed roads,²⁷ whereas in Montreal, Canada, urban environment safety for pedestrians and cyclists is associated with greater neighborhood affluence.²⁸

Two broad categories of factors—individual and contextual—can explain neighborhood inequalities in RTIs.^12,13 Although income and education levels are well-documented individual factors, a recent multilevel analysis demonstrated that the socioeconomic characteristics of individuals and communities exerted independent and additive effects on risk of road death.⁹ Cross-sectional surveys have shown that children from lower income families and those living in downtown areas cross more roads, encounter more motor vehicles every day, and have a higher risk of injury.^29–34

Several ecological investigations have shown the influence of population characteristics and environmental context on the geographical distribution of pedestrians and cyclists injured in urban settings.^13,35–46 In the United States, at the county level, urban sprawl and lower density—which are known to generate more traffic—have been associated with a greater incidence of pedestrian fatalities.³⁵ Within a city, it is generally observed that the per capita or per road kilometer rate of injured pedestrians in a neighborhood increases with population density.^36–41 However, 1 study showed an inverse relationship when other factors were taken into account.⁴² Urban areas with better public transit availability,⁴³ more traffic,^38,41,44 greater density of major roads, or more traffic-generating activities^37,42,45 have a higher incidence of injured pedestrians. In some studies, the proportion of low-income households, the proportion of people without access to a motor vehicle, or an index of multiple deprivation were independent risk factors for pedestrian injury.^{13,39,41,43,46} The ecological design of these investigations precludes conclusions about relationships at the street or intersection level.⁴⁷

At the intersection level, estimates from mathematical models have shown that the mean number of pedestrian crashes is approximately proportional to the square root of vehicle volume.^48–50 The influence of road geometry is also widely acknowledged. Vehicle speed, road crashes, and fatality frequencies are higher on wider major roads (“arteries”), and in general there are more conflict points, crashes, and injuries at 4-legged intersections than at 3-legged, T-intersections.^{22,48,50–52}

Although many studies point to the important role of traffic volume and street geometry in explaining RTIs, few studies have examined the variables explaining socioeconomic inequalities related to RTIs.^{11–13,43,53} When included in multivariate analysis, socioeconomic position is treated as an additional risk factor to be controlled. Therefore, we examined whether differential traffic volume and road geometry can explain socioeconomic variations in the rate of pedestrian and cyclist injuries within an urban area. To improve generalizability to other road users, we also included injured motor vehicle occupants in our analyses.

METHODS

The Island of Montreal was the area of study. In 2001, this 482 square kilometer territory had a population of 1.8 million people and comprised 27 administrative boroughs. The 27 boroughs were subdivided into 521 census tracts, with the mean and median area of census tracts being 0.96 and 0.49 square kilometers, respectively.

The data source for occurrence of RTIs was the sole ambulance service on the Island of Montreal (Urgences-santé Corporation). We included all pedestrians, cyclists, and motor vehicle occupants injured between January 1, 1999, and December 31, 2003, for whom an ambulance was sent.^54,55 Among the 36 763 injured people, we focused on the 21 224 (58%) road users injured at intersections. After excluding complex intersections and census tracts without residents, we retained 3025 pedestrian, 2613 cyclist, and 13 930 motor vehicle occupant reports for analysis. We considered individuals severely injured if they had a prehospital index greater than 3 or the indication of major trauma on prehospital intervention reports.⁵⁶

We mapped geographic location of RTIs onto the road network using ArcInfo version 9.1 (Esri, Redlands, CA). We used the Montreal hierarchical street network (geobase; City of Montreal, Canada) to assign RTIs located within a radius of 15 meters to each intersection.

We derived traffic volume estimates from the Montreal 1998 Origin-Destination survey.⁵⁷ We used data from 65 227 households to construct a representative portrait of trips made by metropolitan area residents during a typical weekday. We used a trip allocation model to identify plausible routes along the road network. We then added up estimates of the average daily traffic volumes on Montreal roads obtained in this way for each intersection to estimate traffic volume at the intersection. These estimates were correlated (r = 0.61) with the estimates from 3-hour traffic counts at 517 intersections provided by the City of Montreal. The analysis included 17 498 intersections, after exclusion of complex intersections—connected to expressway accesses (n = 258) or with traffic volume estimates greater than 80 000 vehicles per day (n = 54)—and of intersections (n = 138) located in census tracts with no residents.

We used the Montreal hierarchical street network to categorize roads as either minor (local streets and collector roads) or major (minor and major arterial roads). A spatial junction of the street network with buffers (15 m) surrounding intersections (ArcInfo v9.1) provided the number of legs per intersection.

Data extracted from the 2001 Canadian Census provided information on household income, population density, and work commute travel modes.⁵⁸ The Origin-Destination survey provided motor vehicle ownership. Because people who use public transit have to walk to and from transit stops,⁵⁹ we pooled these modes of transportation in the multivariate analysis. We used the proportion of people walking or using public transit and the proportion of people cycling to get to work as proxies for pedestrian and cyclist volumes, respectively. We excluded census tracts with no working population (n = 15).

We grouped intersections according to average household income of census tract residents (quintiles) and applied tests for linearity to detect the presence of significant inequality trends in the 3 outcome variables (numbers of injured pedestrians, cyclists, and motor vehicle occupants) and in exposure variables.

We used a multilevel Poisson regression that nested RTIs occurring at 17 498 intersections in the 506 census tracts to determine associations between exposure variables and outcomes (event rate ratios [ERRs]) and to assess the impact of inclusion of other variables on socioeconomic gradients. We used Hierarchical Linear Modeling software version 6.04 (Scientific Software International, Lincolnwood, IL).⁶⁰ We did not include the percentage of census tract households without motor vehicle ownership in the multilevel models because it was strongly correlated with the proportion of residents getting to work by car (r = −0.86) or by public transit or walking (r = 0.85). Traffic volume and the presence of a major road were correlated (point-biserial correlation coefficient r = 0.62). There were correlations between 4-legged intersections and traffic volume (point-biserial correlation coefficient r = 0.26) or the presence of a major road (Φ = 0.25).

We estimated a first null model, which allowed computation of plausible value ranges—or the extent of variation in the number of people injured at intersections across census tracts. The borough's socioeconomic position was the only predictor of the baseline model (model 1), whereas traffic volume (model 2), intersection geometry characteristics (model 3), population density, and modes of transportation used to get to work (model 4) were additional predictors in the more complex models. We modeled traffic volume as a random effect because model comparison tests, derived from respective deviance statistics, showed that inclusion of statistically significant variations (P < .001) in the effect of traffic volume (γ₁₀) across census tracts improved model fit.

We centered the traffic variable at 500 vehicles per day to make the intercept (γ₀₀) easier to interpret. The reference categories for intersections with 1 or more major roads and 4-legged intersections are intersections with only minor roads and 3-legged intersections, respectively. To estimate potential public health gains, we calculated the attributable fraction among the exposed⁶¹ intersections (AF_e = [ERR − 1]/ERR) for the poorest areas.

RESULTS

At intersections in the poorest census tracts, there were on average 6.3 times more pedestrians injured, 3.9 times more cyclists injured, and 4.3 times more motor vehicle occupants injured than in the wealthiest census tracts (Table 1). There was a statistically significant inverse relationship between the household income of populations living in a census tract and the average number of injured individuals and severely injured individuals at intersections.

TABLE 1—

Characteristics of Intersections (n = 17 498) and Census Tracts (n = 506) as a Function of Census Tract Socioeconomic Position: Island of Montreal, Canada, 1999–2003

		Average Household Income of Census Tract Residents, Quintiles
Characteristics of Intersections and Census Tracts	Island of Montreal	Poorest Quintile	Second Poorest Quintile	Middle Quintile	Second Richest Quintile	Richest Quintile	Pa	Poorest to Richest Ratio
Intersections
Injured pedestrians/100 intersections, mean	17.3	35.8	37.3	19.7	13.6	5.7	< .001	6.3
Severely injured pedestrians/100 intersections, mean	3.1	6.9	6.0	3.8	2.2	1.0	< .001	6.6
Youngb pedestrians injured/100 intersections, mean	3.2	7.0	7.2	3.4	2.3	1.0	< .001	7.3
Injured cyclists/100 intersections, mean	14.9	29.0	25.9	18.5	11.2	7.4	< .001	3.9
Severely injured cyclists/100 intersections, mean	1.3	2.7	2.5	1.7	0.9	0.5	< .001	5.2
Youngb cyclists injured/100 intersections, mean	3.3	6.8	5.4	4.1	2.4	1.8	< .001	3.8
Injured MVOs/100 intersections, mean	79.6	143.1	151.1	95.9	69.1	33.5	< .001	4.3
Severely injured MVOs/100 intersections, mean	6.6	13.4	11.4	7.4	5.4	3.3	< .001	4.1
Youngb MVOs injured/100 intersections, mean	5.0	7.7	9.3	6.0	4.7	2.3	< .001	3.3
Traffic at intersection, mean	4748	7651	7350	5120	4065	3189	< .001	2.4
Traffic at intersection, median	1079	1984	1669	1154	1130	653	< .001	3.0
Intersections with ≥ 1 major road, %	17	30	23	18	16	12	< .001	2.6
Four-legged intersections, %	41	51	60	48	38	28	< .001	1.8
Census Tracts
Mean household income of residents, $	50 795	29 741	36 551	42 715	51 870	93 178	< .001	0.3
Population density/km2	7994	10 944	10 164	8472	6380	4003	< .001	2.7
Households without motor vehicle ownership, %	33	49	41	34	25	15	< .001	3.3
Residents on the workforce, transportation to work, %
Walking	9.8	13.5	10.7	9.7	8.0	6.9	< .001	2.0
Cycling	2.0	2.6	2.3	2.2	1.9	0.9	< .001	2.9
Public transit	33.7	43.9	39.9	35.6	29.7	19.5	< .001	2.2
Car	53.6	39.1	46.1	51.5	59.7	71.8	< .001	0.5

Note. MVOs = motor vehicle occupants.

^aTest for linearity.

^bSchool-aged children (5–17 years).

Traffic volume at intersections increased significantly with poverty. At intersections in the poorest census tracts, average and median traffic were, respectively, 2.4 and 3.0 times higher than at intersections in the wealthiest census tracts (Table 1). Arteries were present at 30% of intersections in the poorest census tracts, which is 2.6 times more than in the wealthiest census tracts. There were 1.8 times as many intersections with 4 legs in the poorest census tracts as in the wealthiest census tracts.

Each census tract quintile included 18% to 22% of the population on the Island of Montreal. However, population density increased with poverty, with the poorest census tracts being 2.7 times more densely populated than the wealthiest census tracts (Table 1). Mode of transportation used to get to work also varied according to socioeconomic position. In the poorest census tracts, almost half (49%) of households did not own a motor vehicle, a figure that is 3.3 times higher than in the wealthiest census tracts. Walking, cycling, or using public transit to get to work increased significantly with level of poverty.

The final multilevel models (Table 2) showed that an increase of 1000 vehicles per day at an intersection was associated with a 6% increase in the number of injured pedestrians (ERR = 1.06; 95% confidence interval [CI] = 1.05, 1.06), a 5% increase of injured cyclists (ERR = 1.05; 95% CI = 1.04, 1.06), and a 7% increase of injured motor vehicle occupants (ERR = 1.07; 95% CI = 1.06, 1.08). At intersections with 1 or more major roads, there was a higher number of injured pedestrians (ERR = 2.4; 95% CI = 2.1, 2.7), cyclists (ERR = 1.3; 95% CI = 1.2, 1.5), and motor vehicle occupants (ERR = 3.5; 95% CI = 3.2, 3.8). More pedestrians (ERR = 3.4; 95% CI = 3.1, 3.8), cyclists (ERR = 2.4; 95% CI = 2.2, 2.7), and motor vehicle occupants (ERR = 4.7; 95% CI = 4.4, 5.1) were injured at 4-legged intersections than at 3-legged intersections.

TABLE 2—

Multilevel Multivariate Modeling Analyses Predicting Number of Injured Pedestrians, Cyclists, and Motor Vehicle Occupants: Island of Montreal, Canada, 1999–2003

Exposure Variable	Model 1, ERR (95% CI)	Model 2, ERR (95% CI)	Model 3, ERR (95% CI)	Model 4, ERR (95% CI)
Pedestrians injured, no.
Poorest quintile of mean household income	6.027 (4.462, 8.141)	4.559 (3.431, 6.058)	3.341 (2.599, 4.295)	1.825 (1.341, 2.482)
Fourth quintile	6.103 (4.513, 8.254)	4.465 (3.353, 5.946)	3.068 (2.382, 3.953)	1.936 (1.458, 2.572)
Third quintile	3.489 (2.569, 4.738)	2.771 (2.075, 3.699)	2.244 (1.739, 2.895)	1.570 (1.201, 2.052)
Second quintile	2.320 (1.708, 3.150)	2.077 (1.556, 2.773)	1.854 (1.434, 2.397)	1.489 (1.156, 1.919)
Richest quintile (Ref)	1.000	1.000	1.000	1.000
Traffic volume at intersection (centered at 500 vehicles/d)		1.122 (1.112, 1.132)	1.059 (1.052, 1.067)	1.056 (1.048, 1.063)
Intersection with at least 1 major (vs only minor) road			2.376 (2.131, 2.649)	2.393 (2.145, 2.670)
Intersection with 4 (vs 3) legs			3.525 (3.178, 3.909)	3.442 (3.102, 3.819)
Proportion of residents who walk or use public transit to get to work				1.013 (1.006, 1.020)
Population density/km2				1.038 (1.021, 1.056)
Cyclists injured, no.
Poorest quintile of mean household income	4.273 (3.205, 5.698)	3.538 (2.749, 4.553)	2.985 (2.344, 3.801)	2.389 (1.850, 3.084)
Fourth quintile	3.798 (2.839, 5.082)	3.195 (2.473, 4.129)	2.483 (1.941, 3.177)	2.081 (1.614, 2.684)
Third quintile	2.754 (2.058, 3.685)	2.373 (1.839, 3.061)	2.035 (1.594, 2.596)	1.737 (1.363, 2.215)
Second quintile	1.616 (1.204, 2.168)	1.493 (1.155, 1.931)	1.370 (1.071, 1.752)	1.218 (0.960, 1.545)
Richest quintile (Ref)	1.000	1.000	1.000	1.000
Traffic volume at intersection (centered at 500 vehicle/d)		1.078 (1.070, 1.085)	1.050 (1.043, 1.057)	1.048 (1.041, 1.055)
Intersection with at least 1 major (vs only minor) road			1.301 (1.153, 1.478)	1.324 (1.169, 1.499)
Intersection with 4 (vs 3) legs			2.436 (2.206, 2.688)	2.408 (2.180, 2.660)
Proportion of residents on the workforce that report cycling to get to work				1.100 (1.067, 1.135)
Population density/km2				1.009 (0.992, 1.026)
Motor vehicle occupants injured, no.
Poorest quintile of mean household income	4.257 (3.321, 5.458)	3.458 (2.717, 4.399)	2.384 (1.935, 2.937)
Fourth quintile	4.406 (3.435, 5.651)	3.898 (3.067, 4.954)	2.464 (2.000, 3.035)
Third quintile	2.804 (2.180, 3.608)	2.256 (1.769, 2.877)	1.752 (1.421, 2.160)
Second quintile	2.082 (1.623, 2.672)	2.036 (1.605, 2.582)	1.781 (1.449, 2.188)
Richest quintile (Ref)	1.000	1.000	1.000
Traffic volume at intersection (centered at 500 vehicles/d)		1.162 (1.149, 1.176)	1.067 (1.059, 1.076)
Intersection with at least 1 major road (vs only minor)			3.521 (3.241, 3.826)
Intersection with 4 (vs 3) legs			4.720 (4.354, 5.117)

Note. CI = confidence interval; ERR = event rate ratio. We performed multilevel multivariate modeling analyses predicting number of pedestrians, cyclists, and motor vehicle occupants injured across 17 498 intersections as a function of census tract (household income, population density, commute travel modes) and intersection (traffic volume, road type, number of legs) characteristics. In the multilevel analysis, we nested the 17 498 intersections within the 506 census tracts. Model 1 includes 4 dummy variables for quintiles of household income of census tract residents as the only predictors. Model 2 includes predictors from model 1 and traffic volume at the intersection. Model 3 includes predictors from model 2 and the presence of a major road or 4 legs at the intersection. Model 4 includes predictors from model 3 and population density as well as proportion of residents on the workforce that report walking or cycling to work.

Once we accounted for traffic volume and geometry (arterial roads, number of legs), the number of pedestrians injured at an intersection increased significantly with population density (ERR = 1.04; 95% CI = 1.02, 1.06) and proportion of residents who reported walking or using public transit to get to work (ERR = 1.01; 95% CI = 1.01, 1.02). The number of cyclists injured at an intersection increased significantly with the proportion of residents in the workforce who cycle to get to work (ERR = 1.10; 95% CI = 1.07, 1.14). Other factors taken into account—population density (P ≥ .99) and proportion of people who reported using cars to get to work (P = .45)—were not significantly associated with motor vehicle occupants injured at intersections (results not shown).

The baseline model (model 1), with residents’ average household income as the only predictor, showed that at intersections in the poorest areas there were significantly more injured pedestrians (ERR = 6.0; 95% CI = 4.5, 8.1), cyclists (ERR = 4.3; 95% CI = 3.2, 5.7), and motor vehicle occupants (ERR = 4.3; 95% CI = 3.3, 5.5) than at intersections in the wealthiest areas. Taking into account traffic volume and intersection geometry variables (model 3) greatly attenuated ratios for injured pedestrians (−45%; ERR = 3.3; 95% CI = 2.6, 4.3), cyclists (−30%; ERR = 3.0; 95% CI = 2.3, 3.8), and motor vehicle occupants (−44%; ERR = 2.4; 95% CI = 1.9, 2.9), although for cyclists there was an overlap of 95% CIs (Table 2, Figure 1). The inclusion of population density and proportion of residents who reported walking, using public transit, or cycling to get to work—proxies for pedestrian and cyclist volumes—(model 4) further attenuated the socioeconomic gradients of pedestrian and cyclist injuries. In this final model, the ERRs between intersections in the poorest (fifth quintile) and the richest (first quintile) areas were attenuated to 1.8 (95% CI = 1.3, 2.5) for injured pedestrians and to 2.4 (95% CI = 1.9, 3.1) for injured cyclists, which represent respective reductions of 70% and 44% compared with the ERR in the initial model (Table 2, Figure 1).

FIGURE 1—

Association between socioeconomic position of census tract residents and the mean number of injured pedestrians at intersections in increasingly complex multivariate models including (a) household income only (model 1); (b) household income plus traffic volume (model 2); (c) household income, traffic volume, and intersection geometry (model 3); and (d) household income, traffic volume, intersection geometry, and proxies of pedestrian volume (model 4): Island of Montreal, Canada, 1999–2003.

Note. Richest quintile is the reference group.

Figure 2 shows that, taking into account other factors, similar intersections have a lower number of expected injured road users in the richest area (first quintile) compared with the poorest area (fifth quintile). However, within each census tract quintile, in richer as in poorer areas, the numbers of injured pedestrians, cyclists, and motor vehicle occupants are greater at intersections with 1 or more major roads or a fourth leg.

FIGURE 2—

Predicted number of (a) pedestrians, (b) cyclists, and (c) motor vehicle occupants injured at 4 types of intersections with traffic volumes of 5000 vehicles per day as a function of census tract socioeconomic position: Island of Montreal, Canada, 1999–2003.

Note. The second poorest census tract quintile and second richest census tract quintile are not shown. We held other factors constant, using average values of population density (7994 km²), proportion of residents on the workforce that report walking or using public transit (43.5%) or cycling (2.0%) to get to work. We produced estimates from final multilevel models.

According to the final models, if the average daily traffic at intersections in the poorest census tracts were equal to that in wealthiest census tracts (3189 rather than 7651 vehicles per day), there would be 21% fewer pedestrians, 19% fewer cyclists, and 25% fewer motor vehicle occupants injured at intersections in these areas. If the number of intersections with arterial roads were similar to that in the wealthiest census tracts (12% rather than 30%), at 1 out of every 6 intersections there would be an additional decline in the number of injured pedestrians (−58%), cyclists (−24%), and motor vehicle occupants (−72%). If the number of 4-legged intersections were similar to that in the wealthiest census tracts (28% rather than 51%), at 1 out of every 4 intersections there would be an additional decline in the number of injured pedestrians (−71%), cyclists (−58%), and motor vehicle occupants (−79%).

DISCUSSION

We examined whether socioeconomic variations in the rate of pedestrian and cyclist injuries within an urban area can be explained by differential traffic volume and road geometry. We found that road users in poorer neighborhoods have a higher exposure to traffic and, traffic volume being equal, a greater risk of injury because of the presence of more major roads and 4-legged intersections. However, the results principally show that a substantial portion of the excess rate of RTIs in the poorest urban areas can be explained by the number of people exposed to crashes, traffic volume, and road geometry.

The magnitude of observed inequalities in RTIs across urban neighborhoods is consistent with previous studies. However, most studies were limited to children and to 1 or 2 types of road users—mostly pedestrians—or combined all types of motor vehicle crashes.^{9,37,39,40,42,46,51,62,63} We found that environmental factors associated with a greater risk of crashes were more frequent in the poorest neighborhoods. As in California,^17,18 the poorest neighborhoods in Montreal have more traffic and more major roads. The additional hazard posed by the greater proportion of 4-legged intersections in poorer urban areas has not been reported before, to our knowledge. Greater population density, walking, cycling, and public transit use in the poorest neighborhoods expose more pedestrians and cyclists to potential RTIs. Although we did not take into account all potential risk factors, the attributable fraction estimates suggest that the number of injured pedestrians, cyclists, and motor vehicle occupants would be greatly reduced in the poorest neighborhoods if intersections in these areas were similar to those in the wealthiest neighborhoods.

Despite the independent “effect” of neighborhood socioeconomic position shown in multivariate analyses,^39,41,46 it should be underscored that poverty per se does not produce RTIs—exposure to moving vehicles does.^14,15 The marked attenuation of socioeconomic gradients when traffic volume and road geometry are taken into account (Table 2, Figure 1) illustrates the mediating effect of these variables. Thus, it is plausible to argue that along the causal pathway that leads from neighborhood socioeconomic position to RTIs, the number of people at risk, traffic volume, and road geometry are intermediate variables—or mediators.

The association between neighborhood socioeconomic position and number of people injured at intersections remained statistically significant in the final models. However, the inclusion of more comprehensive environmental variables—associated with pedestrian activity (e.g., public transportation infrastructure) or with the likelihood of crashes (e.g., roadway width, traffic calming)—may further reduce the independent “effect” of neighborhood socioeconomic position to nonsignificance.

Strengths and Limitations

Significant spatial variation in the underreporting of RTIs is possible but unlikely, because in Montreal the sole provider of ambulance services is reached through 911 calls and has a monopoly on ambulance services. In addition, universal health care directly pays transportation and hospitalization fees. A large random travel survey (i.e., the Origin-Destination survey) implemented for transportation planning purposes and covering the whole metropolitan area makes the estimation of traffic possible for all Montreal intersections. Unlike ecological studies relying on area-level aggregated proxies of the local environment, we analyzed the association between the number of RTIs, traffic volume, and road geometry at the intersection level. Unlike usual street-level studies,^48–50,64 our research was not limited to a small sample of sites. According to a population-based approach,^65,66 we included almost all intersections in a large urban area, irrespective of their characteristics or crash history. Focusing on only high-risk or “black spot”^55,67 intersections would have missed the majority of RTIs because 61% of pedestrians and 74% of cyclists were injured at crash sites with only 1 or 2 injured pedestrians or cyclists over the 5-year period.

Population density and mode of transportation used to get to work are common proxies for pedestrian activity at the local level in urban areas,^42,45,51,68 but better estimations of pedestrian activity at the street level for whole urban areas is needed. Without more information on specific intersection design, we cannot attribute the greater risk observed at intersections with arteries to any of the specific features associated with increased traffic capacity. For example, wider arterial roads and lanes may increase operating speeds and stopping distance, whereas increased turning radii and free-flow right turn lanes may specifically jeopardize pedestrians. To better orient street-level environmental preventive strategies, further studies should include more detailed measures of street design. Beyond factors associated with the built environment, we did not include behavioral factors, such as alcohol consumption and seat belt use. These unmeasured environmental and behavioral factors may also contribute to neighborhood RTIs’ inequalities and should be investigated in the future.

Implication for Prevention Strategies

Recent conceptual frameworks^19,22 suggest that the built environment is a distal cause of crashes through the mediating effect of traffic volumes, which are “the primary determinants of crash frequency.”²² A population prevention strategy should attempt “to shift the whole distribution of exposure in a favourable direction,”^65(p431) which implies a paradigm shift in favor of more sustainable transportation that would reduce traffic volumes and prioritize public transit.^69,70 Recently, large reductions in road fatalities in the United States have been attributed to reductions in distances driven.⁷¹ Traditional traffic calming methods are expected to reduce injuries irrespective of the socioeconomic conditions. However, because poor urban areas have the greatest exposure to higher traffic volumes and major roads, they may benefit the most from traffic reduction and road redesign. Furthermore, permanent and substantial reductions of traffic volume would make it easier to redesign urban arteries into safer, more complete streets that address the needs of all road users.⁷²

Conclusions

We have shown that traffic volume, road geometry, and the number of exposed people explain a substantial portion of inequalities in RTIs between poorer and richer neighborhoods. Estimates of the effect of environmental variables are similar across types of road users and across multivariate models. Our results contribute to identifying plausible causal pathways for inequalities in RTIs across neighborhoods. They also suggest that large-scale environmental preventive strategies, such as traffic volume reduction and safer roadway design, may have large public health benefits by reducing crashes.