Speed Limits in São Paulo and the Actions for Road Safety and Air Quality

Abstract

Studies carried out have revealed that every day around three thousand people lose their lives in the world due to traffic accidents and poor air quality. Large cities, with their millions of inhabitants and vehicles, face many problems relating to vehicular traffic. In 2015, the speed limit was modified on several roads in the city of São Paulo, Brazil. However, in 2017, the speed limits were increased again, but not on all previous routes. This study analyzed the impact of this change on the number of accidents and pollutant concentrations, over a period of ten years, comparing the periods before and after the implementation of the measure, using real data collected and provided by the authorities of the city and the state transit and environmental companies, on more than forty routes and two nearby air-quality stations. The results showed a clear reduction in the number of accidents without victims on the roads of the city of São Paulo, starting in 2010. Although the restrictive measures imposed by government officials may have contributed to the decrease in the number of accidents, the number of fatalities has not changed so much. Air pollution has not improved substantially with speed changes, as new speed increases have been linked to new episodes of congestion. The average number of fatalities due to accidents has been increasing since 2010 and accidents are becoming more serious. The application of a general linear statistical model (GLM) estimated the impact of the speed reduction policy in terms of the number of injuries avoided per month: 43.4 and 14.1 on other roads and on the Pinheiros highways, respectively. The results highlight the need for a constant data collection by the authorities in cities with high vehicle traffic. The important temporal time trend in terms of reduction of injuries, but not in terms of fatalities and air quality, shows the need to apply joint public policies, not only speed reduction, but also the use of new technologies and raising drivers’ awareness of the problem.

1. Introduction

Transport systems move people and goods, contribute to the development of national economies and facilitate access to employment and services, improving overall society’s quality of life. However, the traffic of vehicles at high speeds has major adverse impacts in terms of road accidents with the consequent deaths and injuries, as well as material, economic and environmental impacts, such as noise and pollutant emissions [1]. Major risk factors that trigger accidents were identified [2]: lack of use of seat belts and child restraint systems, non-use of helmets, driving under the influence of alcohol and drugs, lack of adequate infrastructure and inadequate or excessive speed [3]. Based on the recommendations of the World Report on road traffic injuries and the Commission for Global Road Safety, it was recommended that new road projects should be as tolerant as practicable, to reduce the consequences of driver errors and failures.

In the period between 2011 and 2020, the World Health Organization (WHO) proclaimed the “First Decade of Action for Road Safety”, for the purpose of stabilizing and reducing the forecast level of road traffic fatalities around the world by increasing activities conducted at the national, regional and global levels [3]. Regarding the period between 2021 and 2030, the WHO proclaimed the “Second Decade of Action for Road Safety”, to reduce the number of traffic accidents by half [4]. It may be observed that the number of traffic accidents remains unacceptably high worldwide: more than 1.35 million people being killed, and up to 50 million injured per year. Developing countries concentrate 90% of all the victims, with this being the leading cause of death worldwide for children and young people between 15 and 29 years of age [4]. Air quality and traffic-related deaths also increased worldwide. Air pollution is among the five biggest causes that contribute to global mortality, accounting for around 4 million deaths per year [5]. Cities with accelerated urban surface growth are deeply impacted by air pollution, leading to a scenario of environmental degradation and unsustainable standards [6].

An important question for the authorities is whether speed limits should be lowered to bring good results to society and the associated cost-benefits. Although the lowering of speed limits implies increases in transportation costs and travel time, it also reduces pollution rates, fuel consumption and accidents [7]. Hosseinlou et al. [8] used a mathematical model to determine the optimal speed for the vehicles depending on location, socio-economic and technological factors. Aarts and Van Schagen [9] reviewed empirical studies and concluded that the crash rate increases with speed increases: vehicles that move much faster than the surrounding traffic have an increasing crash rate. Kroyer et al. [10] developed a risk curve in urban environments, observing that increases in vehicle speed have significant impacts on fatal accident risks (although caution needs to be taken when considering the increase in speed and other parameters). Automotive safety systems designed to avoid accidents and improve road safety are also important [11]. In this sense, Doecke et al. [12] studied the relationship between speed limits and injury severities for different crash types. Results of this study have demonstrated a generally positive exponential relationship between speed limits and fatality rates. Highways permitting motorists to drive more than 100 km/h are only safe from fatal accidents if they have excellent conditions of use, good geometrical projection, and safe roadside design. These reviewed studies found that for a fatality-rate threshold of 1 in 100 crashes, the safe speed limits are the following: 40 km/h for pedestrian crashes, 50 km/h for head-on crashes, 60 km/h for hit fixed-object crashes, 80 km/h for right-angle, right-turn, and left-road/rollover crashes and 110 km/h or more for rear-end crashes [9,10,11,12]. Drivers’ attitudes and behaviors can also influence the occurrence of accidents [13].

The WHO stresses the importance of the local authorities legislating on speed limit reduction [14] at traffic generator hubs attracting large vehicle and pedestrian flows, such as schools and commercial areas. In compliance with the United Nations (UN) recommendations, the city of São Paulo started a life protection program in 2013, with the objective of reducing death rates from traffic accidents to a maximum of three deaths for every 100,000 inhabitants by 2030 [3]. A study carried out in the city of São Paulo concluded that the speed reduction programs applied since 2015 had reduced both the average speed of motor vehicles and the mortality from traffic accidents. The mortality reduction had been more pronounced among people over 50 years old [15,16]. In addition to reducing the number of accidents, lowering the speed limit can also help reduce polluting emissions, fuel consumption and travel times [8]. In studies carried out in Spain, the authors demonstrated that the percentages of reduction of these factors vary, according to the characteristics of each locality [7].

The vehicle fleet in the Metropolitan Region of São Paulo consists of more than 7 million vehicles, including heavy vehicles such as trucks, buses, minibuses, pickup trucks and vans [17]. In 2015, the speed limit was modified on several roads in the city of São Paulo, including those on the Tietê and Pinheiros highways [18]. These highways represent two of the most important roads in the city; they interconnect the main highways, have a large flow of vehicles (with a consequent large emission of pollutants) and concentrate 4.1% of fatal traffic accidents (occupying the first two positions for this type of accident within the area under the responsibility of the municipality). In January 2017, speed limits on the two highways returned to their highest values of 2015. Speed limits remained at more restrictive levels on other roads in the city [19]. A study in São Paulo showed that this speed reduction policy adopted in 2015 led, in its first 18 months of operation, to a 21.7% reduction in the number of accidents, including those with and without fatalities [20]. A similar reduction percentage was previously achieved in the region of Catalonia (Spain), where road speeds were reduced from 120 km/h and 100 km/h to 80 km/h in July 2007. As in São Paulo, the newly elected government in Catalonia revoked this measure in February 2011, allowing a return to previous speed limits and masking impacts of the policy [21].

Our objective in this paper is to analyze whether the reduction in the speed limit on marginal roads in São Paulo contributed to the decrease in the occurrence of traffic accidents, and whether the reduction on other roads in the city, still in force, also had an influence. The study period is between 2010 and 2020, covers the period of the United Nations’ “Decade of Action for Traffic Safety” and uses a 10-year database from São Paulo to assess the impacts of speed reduction and increase policies on traffic safety: frequency of accidents, fatalities, and injuries. Finally, improvements in air quality due to speed reduction strategies are tested on the Pinheiros highway (considered as a case study).

2. Material and Methods

The Tietê marginal has an extension of ~23 km. The highway interconnects the Ayrton Senna and Pinheiros highways, and has three classes of lanes with a road classification composed of fast rapid transit (FT), central lanes (C) and arterial lanes (A). The Pinheiros marginal also has an extension of ~23 km and interconnects the Tietê marginal and other important avenues. This highway has two classes of lanes, with a road classification composed of fast rapid transit and arterial lanes. In the two marginals, speeds have been reduced from 20 July 2015 to 24 January 2017 and increased again since 25 January 2017.

Table 1. Dates and speed limits adopted in each period for the Tietê and Pinheiros highways and for the vehicle types (heavy- and light-duty).

Lane Types	Light-Duty Vehicles	Heavy-Duty Vehicles	Light-Duty Vehicles	Heavy-Duty Vehicles	Light-Duty Vehicles	Heavy-Duty Vehicles
	Until 19 July 2015		From 20 July 2015 to 24 January 2017		Since 25 January 2017
FT 1	90 km/h	70 km/h	70 km/h	60 km/h	90 km/h	60 km/h
C 2	70 km/h	70 km/h	60 km/h	60 km/h	70 km/h	60 km/h
A 3	70 km/h	70 km/h	50 km/h	50 km/h	60 km/h	60-50 km/h

¹ fast traffic, ² central and ³ arterial service lanes.

shows the dates and speed limits adopted in each period for the two highways and for light and heavy-duty vehicles (LDVs and HDVs). Figure 1 shows the layout of the two and their surroundings. The Tietê and Pinheiros marginals have 51 and 26 locations, respectively, with radar surveillances installed in their corresponding segments [18].

In addition, forty other roads in the area under the responsibility of the municipality were studied. These forty local roads have a total length of ~246 km: 19 roads with radar and speed bumps (47.5%), 15 roads with only radar (37.5%) and 6 roads with only speed bumps (15%). In the forty other local roads, speeds have been reduced to 50 km/h in July 2015 and have continued at their reduced level.

2.1. Traffic Accidents

Data on traffic accidents from 2010–2020, including dead and injured, were provided by the São Paulo Traffic Engineering Company (CET) through reports and other requests for information [22]. The data collected from city and state authorities were complete and consistent over the entire study period. There were no gaps or inconsistencies in the data. A General Linear Model (GLM) was fitted to the monthly data on traffic accidents and victims (death and injured), using as independent variables (fixed factors) location, year, month and speed reduction/increase policies (speed scenarios). The combinations of year and month of implementation/removal of the transport policy within each year were used as random factors in the accident model. There were months in the city in which the number of accidents decreased significantly, due to the decrease in speed, especially starting from the year 2015. The regression model estimated the accidents in month t, during the period 2010–2020, using the number of dead and injured on the marginal roads Pinheiros and Tietê and the other forty roads:

Accidents (dead or injured monthly i, location j, month t)_i,j,t = a₀ + a₁ P_t + W_t a₂^w,

(1)

where P_t is a dummy variable ranging from 1 to 0 (1 for months with implementation of a speed reduction policy and 0 for months without such a reduction), reflecting month t related to traffic speed restrictions; a₀ (intercept) and a₁ (slowdown scenario effect) are regression coefficients obtained by ordinary least squares. Time trends at location j were included to control for the effects of transport policies in the data period. Therefore, W_t is a vector of annual records that can impact i accidents because of the implementation of the measures, and a₂^w are the regression coefficients related to the years 2010 to 2020.

2.2. Interactions between Traffic Parameters, Accidents and Air Quality

In January 2017, a technical note appeared advocating for speed variations based on studies of traffic flows (and their respective braking distances for light and heavy vehicles) and levels of service that take into consideration ranges of vehicle density per traffic lane [23]. The level of service is a qualitative assessment of the road traffic operating conditions. The level of service considers speed, travel time, traffic restrictions or interruptions, degree of freedom of maneuver and comfort. Finally, the level of service classifies road segments based on their speed and corresponding traffic volumes [24]. Additionally, this technical note recommended some measures that should be implemented, along with speed increments, especially on marginal roads. The main recommended measures were the improvement of speed regulation and educational/warning signage (such as prohibition of mobile phone use while driving, implementation of pedestrian crossings at junctions, use of variable electronic message boards, increased inspection of motorcycles, expansion of traffic operational equipment on these roads and prohibition of the presence of street vendors). Along with these measures, pavement conditions can also have a strong impact on the number of accidents [25].

The relationships between main traffic parameters, flow F (number of vehicles/h/lane), mean speed s (in km/h) and density D (vehicles/km/lane), are defined according to the following equation:

F = s D,

(2)

The capacity C of a road is the maximum value of the vehicle flow that can circulate through a section with uniform characteristics. This maximum value corresponds to the critical density. Transport policy measures that avoid exceeding the capacity of roads are preferred, to reduce accidents and pollution simultaneously [7,21].

3. Results and Discussion

3.1. Accident Trends

Figure 2 shows the rate of traffic fatalities per 100,000 inhabitants in São Paulo, with the goal of reducing accidents by half between 2010 and 2020, from 12 to 6. It can be observed that, until 2017, there was a considerable reduction to 6.6, almost reaching the target of the United Nations and indicating the effectiveness of the road safety interventions implemented during this period [4]; however, from 2017 onwards, this rate fluctuated, showing the need to continue efforts to reduce accidents [18]. Advancement in vehicle safety and traffic management systems in the coming years are necessary to reach the new target of three deaths per hundred thousand inhabitants by 2030, including measures beyond speed reduction, such as the use of autonomous vehicles, which can be suggested for policy makers.

The analyses presented in Figure 3 separated the data from the Pinheiros and Tietê marginal highways from the other forty roads—between 1st January (2010) and 31st December (2020) of the UN decade of safe actions on road traffic—to identify possible similarities between these pathways. As a result of the reduction in speed limits on several roads in the city in 2015, the number of accidents with injuries and fatalities showed a downward trend, with significative differences between the marginal roads and the other roads. The data in Figure 3 indicate a strong downward trend in the number of injuries on both the marginals and the other roads. This downward trend is not clearly observed regarding fatalities, especially on the Pinheiros highway. For the other forty roads, comparing the periods between 2010 and 2014 and between 2015 and 2020, the total number of accidents, with and without victims, dropped by 53% (

Table 2. Descriptive statistics of accidents with fatalities and injuries on marginal roads and other roads, per year.

Year	Mean Fatalities (±sd) 1	Sum of Fatalities 2	Difference in Fatalities (%) 3	Mean Injuries (±sd) 1	Sum of Injuries 2	Difference in Injuries (%) 3
Pinheiros marginal
2010	0.03 ± 0.18	20	–	1.22 ± 0.56	764	–
2011	0.06 ± 0.28	28	40	1.21 ± 0.58	607	−20.55
2012	0.04 ± 0.20	25	−10.71	1.21 ± 0.58	762	25.54
2013	0.04 ± 0.19	23	−8	1.18 ± 0.50	725	−4.86
2014	0.05 ± 0.23	30	30.43	1.21 ± 1.80	755	4.13
2015	0.05 ± 0.23	18	−40	1.15 ± 0.56	435	−42.38
2016	0.04 ± 0.20	11	−38.89	1.11 ± 0.45	281	−35.4
2017	0.06 ± 0.24	14	27.27	1.19 ± 0.63	270	−3.91
2018	0.09 ± 0.33	22	57.14	1.21 ± 1.08	297	10
2019	0.06 ± 0.24	13	−40.91	1.12 ± 0.62	230	−22.56
2020	0.13 ± 0.34	16	23.08	1.09 ± 0.67	132	−42.61
Total		220			5258
Tiete marginal
2010	0.08 ± 0.30	53	–	1.19 ± 0.69	744	–
2011	0.08 ± 0.29	54	1.89	1.25 ± 0.92	808	8.6
2012	0.07 ± 0.27	47	−12.96	1.19 ± 0.71	785	−2.85
2013	0.06 ± 0.26	38	−19.15	1.18 ± 0.64	774	−1.4
2014	0.07 ± 0.26	38	0	1.16 ± 0.64	645	−16.67
2015	0.07 ± 0.29	28	−26.32	1.15 ± 0.60	441	−31.63
2016	0.07 ± 0.27	15	−46.43	1.19 ± 0.72	259	−41.27
2017	0.09 ± 0.32	20	33.33	1.12 ± 0.87	264	1.93
2018	0.06 ± 0.24	14	−30	1.12 ± 0.54	250	−5.3
2019	0.08 ± 0.28	21	50	1.17 ± 0.68	297	18.8
2020	0.12 ± 0.32	17	−19.05	1.07 ± 0.68	158	−46.8
Total		345			5425
Other forty roads
2010	0.06 ± 0.26	262	–	1.30 ± 0.93	5750	–
2011	0.06 ± 0.27	276	5.34	1.26 ± 0.85	5415	−5.83
2012	0.05 ± 0.24	244	−11.59	1.26 ± 0.83	5846	7.96
2013	0.05 ± 0.22	211	−13.52	1.21 ± 0.70	5270	−9.85
2014	0.06 ± 0.27	263	24.64	1.20 ± 0.83	4870	−7.59
2015	0.04 ± 0.20	142	−46	1.19 ± 0.67	4070	−16.43
2016	0.06 ± 0.25	160	12.68	1.17 ± 0.67	3035	−25.43
2017	0.07 ± 0.27	150	−6.25	1.20 ± 0.70	2598	−14.4
2018	0.08 ± 0.28	155	3.33	1.17 ± 0.65	2348	−9.62
2019	0.09 ± 0.32	213	37.41	1.14 ± 0.68	2595	10.52
2020	0.08 ± 0.29	137	−64.32	1.22 ± 0.73	2088	−19.54
Total		2213			43,885

¹ mean and standard deviation (sd) per day, ² total sum per year and ³ percentage differences between consecutive years for the total number of fatalities and injuries.

). In Figure 3, hourly data were aggregated to monthly values to better capture the trend of the effect of the speed limit changes. Specific road segments were also aggregated at three categories, two marginals and forty other roads, to also identify trend differences between spatial categories. As this is an initial study and the topic is still little explored in São Paulo, future analyses could benefit from higher temporal resolution to better research the effect of speed changes, in the case, for instance, of highway segments in the future establishing variable speeds depending on the time, according to traffic volumes and polluting conditions. The spatial resolution could also be improved by analyzing more specific road segments and/or intersections inside or outside the three aggregated spatial categories. The statistical significance of the trend is represented by the R-square of the six spatial/temporal models.

In a study carried out in Sweden, the authors showed that reducing speed from 50 to 40 km/h reduced the number of accidents with serious injuries by 11% after 1 year, on similar roads to the marginals in São Paulo, with no parking lots and sidewalks [26]. But, as the authors themselves report, there is a complex and unagreed relationship in the study between the decrease in the speed limit and the number of accidents [27]. In the US, there is a similar national trend in road safety [11]. Many factors may contribute to the trend in this study, such as the improvement of vehicle safety technologies. Therefore, there is a strong underlying trend in the data, and linear models can provide a basic understanding of the evolution of traffic safety. However, using only linear models we cannot capture all the impacts of speed changes, as shown by the relative stable number of accidents and injuries over the last five years, in Figure 3. Other methods, such as time series modeling and advanced machine learning techniques, can capture the complex dynamics and underlying trends in road safety data, particularly when assessing the impact of speed changes and other policy interventions over time [27].

Table 2 shows the annual data for accidents with and without fatal victims in the three locations studied. The number of accidents with and without fatal victims decreased in 2015, but fluctuated in subsequent years. In 2019, an increase in accidents on the Tietê marginal and other roads, causing an increase in the number of deaths due to accidents (50% and 37%). In general, the number of injuries and deaths due to traffic accidents show a contradictory situation. Although the accidents with victims have decreased, they have been increasingly serious, and the number of deaths due to accidents has increased, mainly since 2016, on all roads.

Between 2010 and 2020, the 9032 accidents on the marginal roads represent 20% of the total of 44,958 accidents studied in this work (the other forty roads, with 35,926 accidents, represent 80%). Regarding accidents with fatalities, the Tietê marginal registered higher records in the period than the Pinheiros marginal (345 vs. 220), except in 2018 (14 vs. 22). The outbreak of the COVID-19 pandemic in 2020 motivated containment measures which were taken to prevent the spread of the virus, which can have an impact on traffic dynamics in the city of São Paulo. Although the restrictive measures imposed by government officials may have contributed to the decrease in the total number of accidents on main roads, such as the two marginal roads, the number of fatalities has not changed significantly (17 and 16). During this atypical year, speed variations in the same lane, or between different lanes, of the highway may have increased the chances of accidents with serious or fatal victims [28].

The percentage difference between consecutive years in Table 2, both for the total number of fatalities and injuries, highlighted that there was a reduction of 40% and 26% in fatal accidents in 2015, when there was a reduction in speed on the Pinheiros and Tietê marginals, respectively (42% and 32% related to injuries). On the other roads, for the same year, there was a reduction of 46% (fatalities) and 16% (injuries). In January 2017, there was an increase in the speed of marginals, related to an increase of 27% (fatalities) and a reduction of 4% (injuries), and an increase of 33% (fatalities) and 2% (injuries), in the Pinheiros and Tietê marginals, respectively. On the other roads, both fatal accidents (6%) and accidents with injuries (14%) were reduced. This indicates that the speed reduction measure was efficient, and if it had been kept within the marginals, deaths and accidents would probably have been avoided. Most of the other forty roads had decreased their speed since 2015, showing a considerable reduction in the number of accidents; the number of fatalities was similar in the subsequent three-year period (2016 to 2018). However, recorded accidents with injuries showed a significant decrease in the period from 2015 to 2018. A study in Norway evaluated the before and after of a speed reduction scenario (from 80 to 60 km/h), which extended for 5 months. The results showed a 25–35% reduction in accidents with injuries, but the authors concluded that the decrease in accidents may not be only caused by the speed reduction [29].

3.2. Accident Modeling

The application of a general linear statistical model (GLM) shows the impact of speed reduction policies and temporal trends on the monthly number of injuries and deaths.

Table 3. Effects of speed reduction policies and temporal trends on road safety in São Paulo; forecast of injury and death rates during the months of the 2010–2020 period; data for the Pinheiros and Tietê marginals and other roads ¹.

Var (Uni.)	Injuries (#/Month)	Fatalities (#/Month)	Var (Uni.)	Injuries (#/Month)	Fatalities (#/Month)	Var (Uni.)	Injuries (#/Month)	Fatalities (#/Month)
Pinheiros Marginal			Tietê Marginal			Other Roads
Speed reduction policy 2			Speed reduction policy 2			Speed reduction policy 2
	–14.1 ± 5.7 *	−0.5 ± 0.7		–5.7 ± 4.4	–0.1 ± 0.8		–43.4 ± 22.7 *	–2.5 ± 2.8
Time trend			Time trend			Time trend
2010	52.7 ± 4.6 *	0.3 ± 0.6	2010	48.8 ± 3.6 *	3.0 ± 0.6 *	2010	261.8 ± 27.0 *	7.9 ± 3.3 *
2011	39.6 ± 4.6 *	1.0 ± 0.6	2011	54.2 ± 3.6 *	3.1 ± 0.6 *	2011	233.9 ± 27.0 *	9.1 ± 3.3 *
2012	52.5 ± 4.6 *	0.7 ± 0.6	2012	52.2 ± 3.6 *	2.5 ± 0.6 *	2012	269.8 ± 27.0 *	6.4 ± 3.3
2013	49.4 ± 4.6 *	0.6 ± 0.6	2013	51.3 ± 3.6 *	1.7 ± 0.6 *	2013	221.8 ± 27.0 *	3.6 ± 3.3
2014	51.9 ± 4.6 *	1.2 ± 0.6 *	2014	40.6 ± 3.6 *	1.8 ± 0.6 *	2014	188.5 ± 27.0 *	8.0 ± 3.3 *
2015	32.3 ± 5.4 *	0.4 ± 0.7	2015	25.9 ± 4.0 *	0.9 ± 0.7	2015	136.3 ± 21.1 *	−1.3 ± 2.6
2016	26.5 ± 7.3 *	0.1 ± 0.9	2016	14.1 ± 5.7 *	−0.1 ± 1.0	2016	78.9 ± 14.7 *	1.9 ± 1.8
2017	12.7 ± 4.6 *	−0.1 ± 0.6	2017	9.3 ± 3.6 *	0.3 ± 0.7	2017	42.5 ± 14.7 *	1.1 ± 1.8
2018	13.7 ± 4.6 *	0.5 ± 0.6	2018	7.7 ± 3.6 *	−0.2 ± 0.6	2018	21.7 ± 14.7	1.5 ± 1.8
2019	8.2 ± 4.6	−0.3 ± 0.6	2019	11.6 ± 3.6 *	0.3 ± 0.6	2019	42.2 ± 14.7 *	6.3 ± 1.8 *
2020	-	-	2020	-	-	2020	−	-
Regression 3			Regression 3			Regression 3
R2	0.79	-	R2	0.84	-	R2	0.92	-
Mean	34.8	1.5	Mean	39.0	2.6	Mean	331.8	16.7

¹ the table shows the coefficients and standard errors of the regressions for the Pinheiros and Tietê roadways and other roads in the city of São Paulo; a total of 132 observations were considered for each type of road, and the research period was from January 2010 to December 2020 (11 years), including monthly averages (Table 2); ² speed changes were treated as “dummy” variables: speed reductions in the marginals from July 2015 to January 2017 and reductions in “other lanes” from the beginning of September 2015 to the present; * analysis of variance p < 0.01; ³ regression model fitted by ordinary least squares. Average: dependent variable (number of monthly injuries and deaths, 2010–2020).

shows that the regression coefficients related to transport policies and temporal trends were statistically significant for the number of injuries, and there was a reduction on the marginals and the other roads in the region. Table 3 provides detailed information on the statistical significance of the findings, including confidence intervals (expressed as standard errors of the regressions) and p-values for the regression coefficients. The effects of transportation policies have little impact on reducing the timing of fatalities. The regression coefficients of the models in São Paulo confirm the linear trends shown in Figure 3 and the descriptive statistics in Table 2. During the period in which speed limits were reduced, the effects of transportation policies in relation to the temporal reduction in the number of injuries are different, depending on the roads studied: 14.1 fewer injuries per month for the Pinheiros marginal, 5.7 for the Tietê marginal and 43.4 for the other routes (95% confidence intervals and standard errors were estimated). The decreasing values of the regression coefficients over the years of the survey show the growing importance of law enforcement measures. For example, in 2010, the values of the number of monthly injuries (52.7, 48.8 and 261.8) were higher than in the 2020 reference year for the Pinheiros marginal, Tietê marginal, and other roads, respectively.

The reduction of injuries was also made possible by radar surveillance systems in all lanes, even with the continued increases in vehicle flows. The importance of transport policies is reflected by the values of the multiple regression coefficients of GLM (R²). These coefficients improved, compared to the coefficients of the simple linear model of Figure 3, especially in the case of the Pinheiros marginal (27%), going from 0.62 to 0.79 (from 0.74 to 0.84 on the Tietê marginal and from 0.83 to 0.92 on other roads). A significant reduction in the number of deaths per month was also observed in some years in the case of the Tietê marginal (2010–2014) and other roads (2010–2011, 2014 and 2020). The fact of the implementation of an uneven speed-reduction policy in the city—the speed on marginal roads experienced a reduction from July 2015 to January 2017 and on the other roads the speed reduction came into force from the beginning of September 2015, to the present time—can explain the difference in the number of injuries between the marginals and the other roads. If the speed had not been increased again in the two marginals from February 2017, the reduction in the number of injuries and deaths could have been even greater. Therefore, this paper considered the detailed information on how the speed limit changes were implemented and enforced differentially between the marginals and the other forty roads, which affected the impact of speed changes. Keeping the other factors constant, the reduction in speed on the other roads in São Paulo was related to the decrease in fatalities of –2.5 deaths per month (−0.5 and −0.1 deaths per month on the Pinherios and Tietê marginal roads, respectively). In Table 3, the speed reduction policy was treated as a single dummy variable.

Other effects could have been tested in the model, which may influence traffic accidents (and air quality). For example, climatic conditions, road maintenance activities, changes in vehicle technology, and driver behavior were not considered in this study over the period 2010–2020. In this sense, Naik et al. [30] showed that rain and warmer air temperatures can lead to more serious accidents. Another study carried out in Croatia showed that fatal accidents occur more frequently on urban links, at night, with poor visibility, speeding and male drivers [31]. Therefore, fluctuations in traffic volume may also have led to changes in accidents, deaths and injuries. In the municipality of São Paulo, the traffic volume from 2010 to 2019 increased by 15% from 6.3 to 7.3 million vehicles per day [32]. Taking all these changes into account, realistic predictions of accidents, injuries and deaths can be made with greater accuracy. By adding other factors and variables, the assumption relating to accidents and time (hour, month or year) could not be linear, like the regression models applied in this study, in Table 3. Instead of linear models, by incorporating exposure measures and developing time series models, researchers can gain deeper insights into the dynamics of accident or fatality rates over time, identify causal factors, evaluate the impact of interventions and make informed predictions for future road safety outcomes [30]. Advanced machine learning and artificial intelligent techniques can also capture the dynamic nature of traffic accident data. Future research in São Paulo plans to develop this type of non-linear model, over time.

3.3. Pinheiros Marginal Case Study

Collaborating with experts in statistics, transportation engineering, and data science can further enhance the modeling process and its interpretation in road safety studies [27]. Impact factors (speed, vehicle flows, exposure to pollutants and other road factors such as congestion episodes) can be quantified in terms of traffic parameters and relationships on some of the highway segments sampled in this study (Equation (2)). In this sense, Figure 4 and Figure 5 show statistical traffic models on the Pinheiros highway, based on hourly relationships of pairwise traffic parameters, which can be rigorously compared with the corresponding accident data (detailed accidents, deaths and injuries per year and month, at aggregated levels). In the analyzed section of the Pinheiros highway, the Eusébio Matoso bridge (in the direction of Castelo Branco), we can observe that congestion episodes increased during the speed increase scenario (the period from 25 January 2017 to 31 December 2018, Figure 5), compared to the speed reduction scenario (the period from 29 May 2015 to 24 January 2017, Figure 4). More episodes of congestion in this section during the period of increased speed, years 2017 and 2018, were related to the increase in the number of deaths between 27 and 57% (Table 2 and Figure 3), contrary to what was expected initially by the São Paulo City Council.

The worsening of traffic accidents was associated with exposure to pollutant concentrations (NO₂, CO and particle matter smaller than 10 and 2.5 micrograms, PM₁₀/PM_2.5), giving a comprehensive view of air quality impacts. Analyzing the air quality data from the CETESB station located in Pinheiros, next to the traffic segment analyzed on the Eusébio Matoso bridge, we can see that the improvement of NO₂ and CO concentration trends stopped (Figure 6A,B) due to the increase in traffic. For example, in the segment located on the Pinheiros highway, the Eusébio Matoso bridge (in the direction of Interlagos, IT), the average number of cars during working days increased from 6000–6250 vehicles per hour, in each direction, to 6500–6750 (from 2016 to 2017). In this section, car flows exceeded 8000 vehicles per hour in each direction during peak hours (7 a.m. to 10 a.m. and 5 p.m. to 8 p.m.). In this section, increases in vehicle flow and related congestion episodes increased the relationship between the hourly concentration of pollutants (NO_X in μg/m³ and CO in ppm) and traffic (in the number of light-duty vehicle units per hour in each direction). Figure 6C–F show the air concentration and car flow relationships, during the central period of working days (7 a.m.–8 p.m.), for NO_X and CO and for the speed reduction and increase scenarios, respectively. Due to higher air concentrations and vehicle flows in the speed increase scenario, the slope of the regression lines increased (Figure 6E,F), compared to the speed reduction scenario (Figure 6C,D), showing worsening local air quality conditions due to traffic management.

3.4. Study Comparison and Research Needs

The literature review carried out in Table 4 shows the relationship between speed reduction measures and the reduction of accidents, as presented in this paper. All the various national and international studies found an average correlation between speed reduction and accident variation: a reduction of ~25% (±15%) in traffic speed limits leads to a reduction of ~31% (±33%) in traffic accidents. Table 4 shows the results of the 29 reviewed studies, in a snapshot. In a similar meta-analysis study by Elvik et al. [11], the authors found the following speed and accident reduction values: ~6% (±4%) vs. ~27% (±19%). Differences in speed reduction rates are due to differences between legal speed limits and average vehicle speeds. Therefore, reducing road speed limits by 20% leads to a 5% reduction in average speeds [29]. In this study, a reduction of 7% in legal speed limits leads to a reduction of 27% in traffic accidents. Even though the findings of this manuscript are specific to São Paulo, the reduction ranges here may be directly applicable to other cities with different traffic conditions, urban layouts, and policies (such as the case studies presented in Table 4). Overall, ~20 to 30% reduction in average speeds (~5 to 7% reduction in legal speeds) leads to similar reduction in vehicle accidents (and there is a potential for generalizing the results). In Table 4, the reduction in traffic is also related to a decrease in pollutant concentrations of ~17% (±77%), as shown in the Pinheiros case study in this paper (24% reduction in NO_X, Figure 6A). In a study carried out by Pérez-Martínez et al. [33] in São Paulo during the COVID-19 episodes, the authors found a good correlation between the reduction in traffic (39%) and the decrease in pollutant concentrations (15–23%). Long-term trends in air quality in the Metropolitan Region of São Paulo were addressed in another study [34]. Concentrations of CO, NO_X and PM₁₀ in São Paulo decreased yearly by 4, 8 and 9%, respectively (whereas sales of gasoline, ethanol, and diesel increased by 3, 24, and 5%). Over the 2000–2013 period, increases in traffic volumes were partially offset by decreases in pollutant concentrations, due to technical improvements in LDV and HDV engines (LDVs were the major mobile source of CO, whereas HDVs were the major source of NO_X and PM₁₀). Therefore, broader factors beyond speed changes might influence changes in pollutant concentrations (as shown in this study by the wider ranges of the standard deviation compared with the mean: 17 vs. 77%). There are also differential effects of speed limit changes on different types of vehicles and different road types.

According to the reviewed studies in Table 4, the assumption between accidents and time (month or year) might not be linear, like the linear regression models applied in this study, in Table 3. However, considering traffic volumes from 2010 to 2020, if there was only an average increase of 1.5% annually, the fluctuation in traffic volume might not have caused significant changes in accidents, deaths, and injuries. Furthermore, during this period, the slowdown values in the city changed slightly from 1112 to 1170 km per day (5%), showing the small influence of long-term traffic variation on vehicle accidents [22]. Taking these traffic changes into account, realistic predictions of accidents or deaths can be made accurately. Instead of the linear models in Table 4, we are planning to develop time series models considering traffic exposure and accident rates expressed in terms of number of deaths and injuries per vehicle kilometer. Non-linear models, over time, can quantify the real impact of various factors: speed, exposure, and other traffic parameters on roads. As such, we need to plan some rigorous statistical models with detailed data on accidents (not just accidents, deaths and injuries per year, at aggregate levels). This study used aggregated data for specific highways and roads, which might mask important patterns. The study could benefit from higher resolutions (i.e., hourly and quarter-hourly data) to better capture the effect of speed limit changes and policies. More granular data, spatially disaggregated by road segments and intersections, could also provide insights into specific factors contributing to accidents and pollution, such as vehicle and road types. The statistical techniques applied in this manuscript can be compared to other statistical models, such as deep learning techniques and global sensitivity analyses, to better capture the dynamic nature of the city’s traffic and congestion data. In this sense, other studies discussed advanced accident modeling in car fatalities [35,36]. Other research attempted to improve monitoring of speed and traffic pollution by using remote sensing monitors and low-cost sensors [37,38].

As shown in Figure 3, there is an overall downward trend in terms of accidents, injuries, and deaths, as in most of the literature reviewed in Table 4. In these studies, there is a similar international trend regarding traffic pollution (mainly in terms of NO_X, CO and PM). Including other pollutants, such as toxic elements, black carbon and nanoparticles, could provide a more comprehensive view of air quality impacts caused. Specific analyses of pollutants and their health impacts is needed, but was beyond the scope of this study. Many factors can contribute to this trend, such as technical improvement of vehicle safety, adequate road maintenance activities, driver behavior changes and traffic management systems [39]. According to the studies reviewed, there is also a relatively stable number of accidents and injuries in the last five or seven years, and there is a strong underlying trend in the data on accidents and pollution. Simply using linear models cannot capture the impact of factors such as speed changes and policies, in this study. It is recommended that we consider time series modeling, especially during the years following the COVID-19 pandemic, to test changes in traffic behaviors and impacts on new technologies.

Table 4. Studies included in the analyses of the relationships between traffic parameters, accidents, and pollution variations.

Ref.	Transport Policy Study/ Measures and Highlights	Time Span	Speed Reduction	Traffic Variation	Accident Variation	Pollution Variation	Pollutant Type
[7]	Speed reduction and pollution in Madrid (Spain)	2011–2017	from 120 to 110 km/h on rural roads	−15%	−20%	−18%	Fuel consumption and CO
			from 90 to 70 km/h on urban roads		−20%
[9]	Speed reduction policy on car-crash accidents	2006	from 60 to 46 km/h on urban roads		−60%
			from 100 to 76 km/h on rural roads		−20%
[20]	Speed reduction policy on car-crash accidents	2015–2016	from 90 to 70 km/h		−22%	−23%	PM2.5
						−35%	CO
[40]	Influence of traffic on PM in Madrid	1999–2001			−30%	−20%	PM2.5–PM10
[21]	Speed reduction policy in Barcelona	2008–2009	from 100–120 to 80 km/h	−6%	−7%	−5.6%	PM10
						−2.5%	NOX
[27]	Relationship between speed and traffic fatalities in US	1987–1995	from 100 to 85 km/h on rural interstate roads	no traffic variation	−21%
[41]	Weather, air pollution and traffic accidents in Taipei (Taiwan)	2018			−28%	−13%	PM2.5
[42]	Mitigation measures, PM2.5 in Beijing (Olympics)	2007-2011		−50%		−16%	PM2.5
[43]	Emission-reduction measures during red air-pollution alert in Beijing (China)	2015	Emission-reduction measures and traffic restrictions	−28%		−15%	PM2.5
[28]	Impact of speed variations on freeway crashes in UK	2017	from 80 to 60 km/h	−10%	−8%
				−25%	−20%
[13]	Relationship between speeding and crashes in British Columbia (Canada)	1985–1990	speed reduction from maximum speed limits	−22%	−31%
					−37%
[44]	Weather effect on air pollution and traffic in Khuzestan State (Iran)	2008–2015		_	−65.4%	−25%	NOX
				_	−5.0%	−25%	NO2
[12]	Optimal speed limits to reduce car accidents in Australia	2000–2014	from 80 to 50 km/h from 110 to 80 km/h	−38% −27%	−90% −64%		_
[11]	Speed reduction policy and Metanalyses in Oslo (Norway)	2004–2005	from 80 to 60 km/h from 90 to 40 km/h	−25% −56%	−67% −94%
[45]	Teleworking effects on cities in Switzerland	2002–2013		−3%		−3%	NO2
				−3%		−4%	CO
[46]	Air pollution alerts and respiratory diseases in South Korea	2015–2019				−8%	PM2.5
[8]	Speed optimization to reduce road accidents and pollution in Shiraz (Iran)	2011	from 82 to 72 km/h		−10%	−4%	Several pollutants
[47]	Traffic-related pollution in Danish cities (Copenhagen/Roskilde)	2005	from 80 to 40 km/h	−36%		−19%	NO2
						−6%	dB
[10]	Accident analyses worldwide	2009–2011	from 70 to 50 km/h	−25%	−62%
			from 40 to 36 km/h	-15%	−30%
[48]	Weather effect on air pollution and traffic in Madrid (Spain)	2006		−27%		−6%	PM10
[33]	Traffic and pollution relationships in São Paulo (Brazil) during COVID-19 lockdown	2019–2020		−39%		−15%	PM2.5
				−39%		−23%	CO
				−39%		−15%	NO2
[49]	Influence of road traffic emissions on air quality in Barcelona (Spain)	1999–2007		−14%		−42%	PM1
[26]	Speed reduction effect on car accidents in Sweden	2014–2015	from 50 to 40 km/h	−4%	−11%
[50]	Reduction in residential speed limits and traffic behavior in Edmonton (Canada)	2004–2009	from 50 to 40 km/h	−9%	−14%
[51]	Health effects for PM2.5 emission reductions in Beijing	2017		−26%		−32%	PM2.5
[52]	War conflict, reduction in traffic volumes and urban pollution in Israel	2005–2006	traffic reduction due to socioeconomic conditions	−40%		−38.5	NO2
[31]	Risk factors in urban accidents in Zagreb (Croatia)	1999–2000	increase from upper speed limits	_	65%
[53]	Air pollution in Beijing	1998–2013				-37%	PM10
This study	Reduction in speed limits at marginal roads in São Paulo	2015–2019	from 90 to 70 km/h	−7%	−27%	−24%	NOX

Table 4. Studies included in the analyses of the relationships between traffic parameters, accidents, and pollution variations.

Ref.	Transport Policy Study/ Measures and Highlights	Time Span	Speed Reduction	Traffic Variation	Accident Variation	Pollution Variation	Pollutant Type
[7]	Speed reduction and pollution in Madrid (Spain)	2011–2017	from 120 to 110 km/h on rural roads	−15%	−20%	−18%	Fuel consumption and CO
			from 90 to 70 km/h on urban roads		−20%
[9]	Speed reduction policy on car-crash accidents	2006	from 60 to 46 km/h on urban roads		−60%
			from 100 to 76 km/h on rural roads		−20%
[20]	Speed reduction policy on car-crash accidents	2015–2016	from 90 to 70 km/h		−22%	−23%	PM2.5
						−35%	CO
[40]	Influence of traffic on PM in Madrid	1999–2001			−30%	−20%	PM2.5–PM10
[21]	Speed reduction policy in Barcelona	2008–2009	from 100–120 to 80 km/h	−6%	−7%	−5.6%	PM10
						−2.5%	NOX
[27]	Relationship between speed and traffic fatalities in US	1987–1995	from 100 to 85 km/h on rural interstate roads	no traffic variation	−21%
[41]	Weather, air pollution and traffic accidents in Taipei (Taiwan)	2018			−28%	−13%	PM2.5
[42]	Mitigation measures, PM2.5 in Beijing (Olympics)	2007-2011		−50%		−16%	PM2.5
[43]	Emission-reduction measures during red air-pollution alert in Beijing (China)	2015	Emission-reduction measures and traffic restrictions	−28%		−15%	PM2.5
[28]	Impact of speed variations on freeway crashes in UK	2017	from 80 to 60 km/h	−10%	−8%
				−25%	−20%
[13]	Relationship between speeding and crashes in British Columbia (Canada)	1985–1990	speed reduction from maximum speed limits	−22%	−31%
					−37%
[44]	Weather effect on air pollution and traffic in Khuzestan State (Iran)	2008–2015		_	−65.4%	−25%	NOX
				_	−5.0%	−25%	NO2
[12]	Optimal speed limits to reduce car accidents in Australia	2000–2014	from 80 to 50 km/h from 110 to 80 km/h	−38% −27%	−90% −64%		_
[11]	Speed reduction policy and Metanalyses in Oslo (Norway)	2004–2005	from 80 to 60 km/h from 90 to 40 km/h	−25% −56%	−67% −94%
[45]	Teleworking effects on cities in Switzerland	2002–2013		−3%		−3%	NO2
				−3%		−4%	CO
[46]	Air pollution alerts and respiratory diseases in South Korea	2015–2019				−8%	PM2.5
[8]	Speed optimization to reduce road accidents and pollution in Shiraz (Iran)	2011	from 82 to 72 km/h		−10%	−4%	Several pollutants
[47]	Traffic-related pollution in Danish cities (Copenhagen/Roskilde)	2005	from 80 to 40 km/h	−36%		−19%	NO2
						−6%	dB
[10]	Accident analyses worldwide	2009–2011	from 70 to 50 km/h	−25%	−62%
			from 40 to 36 km/h	-15%	−30%
[48]	Weather effect on air pollution and traffic in Madrid (Spain)	2006		−27%		−6%	PM10
[33]	Traffic and pollution relationships in São Paulo (Brazil) during COVID-19 lockdown	2019–2020		−39%		−15%	PM2.5
				−39%		−23%	CO
				−39%		−15%	NO2
[49]	Influence of road traffic emissions on air quality in Barcelona (Spain)	1999–2007		−14%		−42%	PM1
[26]	Speed reduction effect on car accidents in Sweden	2014–2015	from 50 to 40 km/h	−4%	−11%
[50]	Reduction in residential speed limits and traffic behavior in Edmonton (Canada)	2004–2009	from 50 to 40 km/h	−9%	−14%
[51]	Health effects for PM2.5 emission reductions in Beijing	2017		−26%		−32%	PM2.5
[52]	War conflict, reduction in traffic volumes and urban pollution in Israel	2005–2006	traffic reduction due to socioeconomic conditions	−40%		−38.5	NO2
[31]	Risk factors in urban accidents in Zagreb (Croatia)	1999–2000	increase from upper speed limits	_	65%
[53]	Air pollution in Beijing	1998–2013				-37%	PM10
This study	Reduction in speed limits at marginal roads in São Paulo	2015–2019	from 90 to 70 km/h	−7%	−27%	−24%	NOX

4. Conclusions

The results in this study show a clear reduction in the number of accidents without victims on the roads of the city of São Paulo, starting in 2010. There was a reduction in accidents on both marginal highways and other roads. However, when there was an increase in speed on the marginal, the relevant numbers began to move away. There was an increase in the total number of accidents, including injuries and deaths, on the Pinheiros and Tietê marginals. The inconsistency of transport policies regarding speed limits explains the main reason for the unchanged fatality rates, despite the reduced accidents. A factor that may also have contributed to the increase in accidents would be the difference between the speed limits for light and heavy vehicles in the same lane and between lanes, since heavy vehicles circulate in the right lane, but this lane is also used by other vehicles (Table 1). The increase in speed also had an impact on air quality worsening in the Pinheiros marginal road (Figure 6).

The results of the model showed that reducing speed on other roads in São Paulo caused a decrease of 30 fatalities per year. Considering that the victims in the marginal roads accounted for ~20% of the fatalities in the period studied, ~6 deaths could have been avoided annually in the marginal roads. When there was an increase in speed on marginal roads, studies have shown that travel time decreased by 5.5% [20]. However, when traffic accidents occur, the road is blocked, due to the rapid assistance of rescue teams. The shorter travel time, which is the main justification for increasing speed, is not advantageous in these situations, and the consequences of the accident can include economic losses, injuries, and lost lives. The number of accidents on marginal roads remains high, and the problem may be related to the legal speed limit, which is incompatible with WHO and safety guidelines. Increasing speed limits can lead to more congestion events, increasing pollutant concentrations, as shown in Figure 4 and Figure 5.

It can be said that the policy of reducing the speed limit in the city, together with the rest of the measures adopted, came close to meeting the objective of saving lives, meeting the Sustainable Development Goal (SDG) 3.6 of halving the number of deaths from road accidents in 2020 (Figure 2). It is necessary to adopt new policies, such as the review of the speed limit on marginal roads, through technical and scientific knowledge, with the participation of civil society through road education campaigns [31,51]. Suggested transport measures for policy makers beyond speed reductions could be investments in advanced vehicle safety technologies and traffic management systems. The use of autonomous vehicles can impact traffic operations and speed–flow relationships, achieving a much more efficient use of existing roadway spaces and reducing accidents and pollutant emissions [39]. It is still necessary to achieve the objective of reducing the number of deaths from traffic accidents by 2030, to a maximum of three per 100,000 inhabitants (Figure 2). The last report released in 2021 by the traffic engineering company [54] shows that the Pinheiros and Tietê marginal highways continue to be the ones with the most accidents in the city, reinforcing the importance of future studies like this one and considering more variables, to raise awareness among decision-makers. Investments in urban tolls, bicycle lanes, better quality of public transport, better road conditions and traffic education campaigns are all measures that can improve traffic (and air quality) in large cities such as São Paulo. Policymakers could create synergies between air quality improvements and road safety enhancements, achieving significant progress in both areas.