The effect of Rodizio Circular in Sao Paulo on air quality and household decisions

(1)

The Effect of Rodizio Circular in Sao Paulo

On Air Quality and Household Decisions.

Brendan Curran

MSc. Economics (Public Policy) University of Amsterdam

August 2017

Prof. Dr. E.J.S. Plug Supervisor

(2)

1 Abstract

In this analysis, the aim is to assess the causal relationship between the Rodizio Circular policy in Sao Paulo and an impact on three air pollutants – Ozone, particle matter greater than 2.5mm and Nitrogen Oxides. This thesis uses data from CETESB (Sao Paulo’s Environmental Protection Agency) to assess the impact across a period of 2013 to 2017. Using fixed effects modelling, with the monitoring station as the entity variable, the results demonstrated a mixed impact on different pollutants indicating the policy wasn’t reducing congestion or pollution. Further evidence on trends across years and weeks suggested that households have been substituting driving intertemporally and review of car ownership supported hypothesis that households were buying additional vehicles.

(3)

2 Statement of Originality

This document is written by student, Brendan Curran, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not the contents.

(4)

3

List of Figures

Figure 1: Details of the Restriction of Rodizio Circular Figure 2: Area of Rodizio Circular in Sao Paulo

Figure 3: Simplified Relationship of NOx Emission Sources and their Formation of Pm2.5 and

O3.

Figure 4: O3 Formation – VOC- Limited and NOx-Limited

Figure 5: Hourly Average Concentrations of Pollutants: NOx & O3.

Figure 6: Pollutant Year on Year Hour of the Day Averages Figure 7: Average Level of Pollution by Day of the Week

Figure 8: Population of Private Cars Owned in Sao Paulo (2008-2017)

Figure 9: Population of Private Motorcycles/ Mopeds Owned in Sao Paulo (2008-2017)

List of Tables

Table I: Summary Statistics Table II: Descriptive Statistics

Table III: Impact of Rodizio Circular Driving Restrictions on O3 Pollution Levels

Table IV: Impact of Rodizio Circular Driving Restrictions on Pm2.5 Pollution Levels

Table V: Impact of Rodizio Circular Driving Restrictions on NOx Pollution Levels Table VI: Robustness Tests of the Results

(6)

5

1.Introduction

1.1 Motivation

Individual countries have been battling with air pollution for centuries. In 14th century England,

regulations were put in place against those burning sea coal along with strong enforcement

mechanisms.1(National Research Council, 2008) However, with centuries of industrialisation

in Western Europe, growing populations and mass use of fossil fuels for heating and transportation, ambient levels of pollution are so high that they are creating health endemics. The World Health Organisation’s (WHO) Air Quality Model has shown that 92% of the

world’s population live in locations where the air pollution levels exceed their own “WHO’s

Ambient Air quality guidelines”2_{for annual mean of particulate matter with a diameter of less}

than 2.5 micrometres (Pm2.5) (WHO, 2005). Pm2.5 is the most damaging airborne particle to

health and hence, ever-increasing levels of this pollutant are having negative impacts on the well-being of the exposed populations (WHO, 2005). In particular, cities are most affected by these high concentrations as this is where people, vehicles and production are most concentrated. Hence, in this thesis the focus will be on the public policy interventions in an urban setting. The aim being to understand whether government intervention is having the desired impact, in terms of reducing pollutants, or whether it drives alternate behaviour from the impacted population (that could, in fact, exacerbate the issue).

1.2 Research question

The policy of focus is Rodizio Circular implemented in the city of Sao Paulo from 1997 to present. Rodizio Circular involves banning vehicles within the city centre, on certain days of the week, at certain hours, based on their license plate number. At least one day a week, an individual car will be banned from driving in the centre. Using data from 2013 to 2017, the

focus is on three main pollutants. These are Ozone (O3), particle matter greater than 2.5mm

(Pm2.5) and Nitrogen Oxides (NOx). This thesis uses fixed effects modelling to assess the impact of the policy. The results show that the policy does not have the intended outcome as it is not associated with a fall in ambient air pollution. During the hours of the restriction there is evidence of a fall in some pollutants, however, in total pollution levels have not fallen due to the policy. Any positive results from the policy are negated by behavioural responses of

1_{In fact, those that did not comply were often subjected to torture highlighting how seriously the issue was} taken.

2_{WHO guideline limits for annual mean of PM}

(7)

6 individual households. These behavioural responses include changes in household behaviour as a result of the policy e.g. buying an additional car or driving at different time to avoid the driving restrictions. In some cases, due to these changes in household behaviour, the policy can even lead to higher concentrations of pollutants.

1.3 Layout of the Thesis

This thesis is divided into six sections. Section 2 offers context on the policy of Rodizio Circular and similar policies. It also gives details on the different pollutants and the associated health costs. Section 3 is a review of the related literature on measuring air pollution and other road rationing schemes. Section 4 provides a description of the data, as well as the theoretical and empirical models employed. Section 5 presents an analytical and graphical review of the results. Tests of robustness are also included in this section. Section 6 concludes and discusses potential policy implications.

2. Background & Context

2.1 Vehicle Emissions and Road Rationing Schemes.

Motor vehicles contribute to various pollutant populations in the air. These include carbon

monoxide (CO), nitrogen oxides (NOx = NO and NO2), ground level formation of ozone (O3)

and particle matter of diameter greater than 2.5mm (Pm2.5)3. There have been various public

policy interventions over the years to try and curb these pollutants. However, the focus of this thesis is road rationing (driving restriction) schemes i.e. where the government provides incentives or deterrents to reduce the actual number of vehicles on the road at certain times. These schemes have been used in recent decades as a policy tool for reducing congestion as well as air pollution. Examples in Europe include the congestion charge in London or the Low Emissions Zones (LEZs) in Germany and Sweden (Holman et al, 2015). In South America, Bogota has introduced numerous different driving restriction policies under the name “Pico y Plata” and Mexico City was a trailblazer of such schemes with the “Hoy No Circular” scheme introduced in 1989. Both Bogota’s and Mexico City’s schemes involved banning certain cars, based on their number plates, from driving in certain areas of the city. This thesis will evaluate a similar scheme called “Rodizio Circular” in Sao Paulo

3_{Pm2.5 constitutes coarse dust particle matter greater than 2.5mm. It can be naturally made but within cities it is} mainly constructed as a result of human activity (WHO, 2005).

(8)

7

2.2 Rodizio Circular in Sao Paulo

Rodizio circular only takes effect on weekdays between the hours of 7am and 10 am and in the

evening between 5pm and 8pm4. Certain vehicles are banned from driving within the Rodizio

Circular area on certain weekdays within these times. The restriction is based on the car’s license plate numbers. Each day alternates two numbers and those vehicles with license plates ending in these numbers are restricted from driving in central Sao Paulo (see schedule in Figure 1). These time scales are aimed at reducing the number of cars and hence, vehicle pollution emitted at key rush hour times. Theoretically, vehicles and emissions should be reduced by 40%. The area of Rodizio Circular (shown in Figure 2) is regulated with cameras on the entrance roads. Furthermore, any driver caught in violation must pay a fine and receives 4 points on their license. Therefore, if a driver is caught 5 times in violation of the law, they can

be banned from driving for up to 24 months.5 Vehicles exempted from the restriction include

buses and other urban transportation vehicles, school buses, ambulances and other medical services vehicles. Hence, a potential consequence is to encourage alternate modes of transport.

Figure 1: Details of Cars Restricted in Rodizio Circular

Day of the Week Forbidden Digits

Monday 1 and 2

Tuesday 3 and 4

Wednesday 5 and 6

Thursday 7 and 8

Friday 9 and 0

Source: Secretaria de Mobilidade e Transport Website6

4_{Full details of the rodizio circular driving restrictions can be found on the governmental website of Sao Paulo} (http://www.governoaberto.sp.gov.br/)

5_{Drivers with more than 19 points are banned for up to 24 months.}

http://www.cetsp.com.br/consultas/rodizio-municipal/como-funciona.aspx

6_{Municipal Secretary of Mobility and Transport}

(9)

8 Figure 2: Area of Rodizio Circular in Sao Paulo

Source: Secretaria de Mobilidade e Transport Website7 2.3 Pollutant Heterogeneity

In understanding the concentrations of different pollutants, a small amount of cross-disciplinary context is required in the field of atmospheric chemistry. The focus is how the

policy impacts O3, NOx and Pm2.5 air pollution.

Fig 3: Simplified Relationship of Nitrogen Oxides (NOx) Emission Sources and their Formation of Pm2.5 and Ozone (O3)

Nitrogen Oxides (NOx) do have negative impact on health individually but also contribute to

the formation of O3 and Pm2.5 (as seen in Figure 3). Nitrogen Oxides are a family of poisonous,

7_{Municipal Secretary of Mobility and Transport}

(10)

9 highly reactive gases. They form when fuel is burnt at high temperatures. They are emitted by automobiles, trucks and various non-road vehicles as well as industrial production e.g. power plants. However, since most of Sao Paulo’s industry became decentralised from the 1980s, vehicles are the main contributor to this pollutant count (World Bank 1983).

Ground-level Ozone (O3) forms photochemically in the troposphere8, whereby sunlight triggers

reactions involving the interaction with Nitrogen Oxides (NOx) and Volatile Organic

Compounds9 (VOCs). Hence, the formation of O3 involves NOx and is highest when the

sunlight (or temperature) is greatest (Stedman, 2004). Therefore, O3 forms at the greatest rate

during the day. This is important to note when designing and evaluating air pollution policy. Seasonal variation and time dependent factors must be taken account of when measuring any

treatment on O3 formation. These precursors -VOC and NOx- to O3 formation have varied

sources. However, in the state of Sao Paulo the main factor is vehicle emissions. For example, automobile emissions account for 70.8% of VOCs. Hence, a rise in vehicle emissions causes

greater ozone formation within the city (Salvo, 2017). The rate of formation of O3 also depends

on whether an area’s atmosphere is VOC-limited or NOx- limited (Jacob, 1999). An

atmosphere that is VOC-limited will result in O3 concentrations increasing in VOCs and

decreasing in NOx. Conversely, an atmosphere that is limited in NOx will be decreasing in VOCs and increasing NOx. Understanding of this terminology is important in interpretation of

this thesis’s results. It is also important for policy makers when designing effective O3 pollution

control policies.

Figure 4: O3 Formation – VOC & NOx Limited

8_{The lowest region of the atmosphere, extending from the earth's surface to a height of about 6–10 km (the}

lower boundary of the stratosphere).

9_{Volatile Organic Compounds (VOCs) are organic compounds that easily become vapours or gases. Along with} carbon, they contain elements such as hydrogen, oxygen, fluorine, bromine, sulphur or nitrogen.

(11)

10

Airborne particle matter (Pm2.5) represents a complex mixture of organic and inorganic

substances. While it can form naturally, human intervention is a more common cause. A review

of the composition of particle matter in Sao Paulo found that Pm2.5 was primarily caused by

vehicle emissions (Souza et al, 2014). Furthermore, it suggested that Pm2.5 formation was

influenced by humidity and temperature also.10

2.4. Health Issues caused by Pollutants

Pm2.5 - A recent WHO report stated that there were 3 million premature deaths annually due to

excessive levels of air pollution (WHO, 2016). The health burden being felt greatest in low and middle-income countries (which includes Brazil), where 88% of these premature deaths occurred. The severity of the health issues was highlighted in a 2016 health assessment by Miraglia and Abe. These researchers reviewed the impact of the high levels of ambient air pollution from 2009 to 2011 on the health of Sao Paulo residents. They found that if Sao Paulo

diminished the level of PM2.5 by even 5 ug/m3, nearly 1724 deaths could be avoided creating

a gain of $4.96bn annually for the city region. The reduction in life expectancy due to long

term exposure to Pm2.5 is mostly due to increased cardio-pulmonary disease and reduced lung

function in adults. (WHO, 2003)

Ozone (O3) - As noted in the previous section, Ozone is not directly emitted but rather is

formed due to photochemical reactions. Nonetheless, it can have serious impact on health even at relatively low levels. With children and the elderly most at risk, it can aggravate asthma, emphysema and chronic bronchitis (Gryparis et al, 2004). Furthermore, it can cause long term damage to lungs, sometimes resulting in chronic pulmonary disease.

Nitrogen Oxides (NOx) - Nitrogen Oxides include both Nitrogen Oxide (NO) and nitrogen

dioxide (NO2). Both are key in the photochemical formation of Ozone (O3) and contribute to

Pm2.5 formation. Nitrogen Dioxide (part of NOx) at high levels can cause inflammation of the

lining of the lungs and aggravate many respiratory-based conditions. However, the pollutant impact is best viewed as a contributory pollutant to the formation of more impactful pollutants (like O3).

10_{Further overview of the formation of the pollutant can be found in the basic textbook, Introduction to} Atmospheric Chemistry, Jacob, Daniel J, 1999)

(12)

11

3. Related Literature

In this thesis, I am concerned with measuring air pollution in cities and the treatment effect of driving restriction policies on air pollution. There have been several papers measuring the effect of pollution levels. Only a few researchers, however, have assessed the effectiveness of road rationing policies such as Rodizio Circular.

3.1 Measuring Air Quality Policy Effectiveness- Road Pricing & LEZs

Road pricing policies are often used as s preferred policy tool to reduce congestion and air pollution in an urban setting. Gibson and Carnovale (2015) exploited a natural experiment created by an unanticipated court injunction to evaluate driver’s responses to road pricing. They found that air pollution and congestion in Milan was reduced substantially within the priced area. Furthermore, the substitution to other forms of transit depended on the availability of public transport.

Wolff (2014) assessed the effectiveness of Low Emission Zones (LEZs) in Europe on air pollution. These policies involve banning of the more ‘dirty’11_{cars from defined areas to} improve air quality. He found, using a difference-in-difference model, that these areas were associated with a 9% fall in air pollution. Wolff highlighted that this was driven by household substitution to cleaner technologies. This substitution increased as they lived closer to an LEZ. This household behavioural response created additional benefits rather than the associated detrimental household response with road rationing schemes.

3.2 Driving Restriction Papers

This paper looks at non-price related control policies i.e. rationing of the roads. As stated previously, there have been a number of road rationing policies, particularly across Latin America. Moreover, in the last decade, there have been some attempts to empirically review their impact on pollution. The ‘Hoy No Circular’ Scheme, within Mexico City metropolitan area, was first assessed by Eskeland and Feyzioglu (1997). They were not primarily interested in the impact on air pollution but rather why the scheme had failed to bring down congestion. In fact, they found that it had increased driving within the capital. They emphasised this “lesson in welfare economics” (Eskeland & Feyzioglu, 1997, 283) could be explained by the substituting behaviours of households. Whether this substituting behaviour was to buy additional cars or drive more on the weekends. They used OLS regression to assess demand

(13)

12 functions for gasoline before and after the policy was implemented and found that demand had risen. The authors produced a discrete choice model using household level data to highlight that purchases of vehicles had risen. Davis (2008) went further and assessed the impact of HNC on hourly means of air pollution between the years of 1986 to 1993. This period spans the policy inception so Davis could use a Regression Discontinuity Design (RDD) to isolate the effect of the policy on air pollutants. Nonetheless, similar to Eskeland and Feyzioglu, he found the driving restriction did not have the intended effect as air pollution was not reduced. This was the first paper to use RDD to assess the immediate impact of the restriction. Gallego et al (2013) took a different approach to measuring the HNC policy. They assessed the impact on Carbon Monoxide (CO) concentration during peak hours, before and after the policy was implemented, using a flexible fit estimation with treatment dummies. Their results found a 13% fall the day after the policy and an 11% increase in the medium-term (12-15 months). This increase in CO was shown to be due to households buying additional cars. They also found household responses to the policy were differentiated along income lines.

However, there is some argument that behavioural adjustments to such policies can be lagged and the window of four years post does not capture this. Most recently, Lawell, Umanskaya and Zhang (2017) used the distinct example of Bogota to assess the impact of slightly varying road restrictions. They used a simple neoclassical household utility to predict potential avenues that households may change their behaviour due to the license plate based restriction. Using empirical evidence from the “Pico y Plata” driving restrictions, they found that, in some instances, pollution actually increased under the restriction. This suggestive evidence was based on running RDDs on the policy implementation and three different changes in the policy. They also found that the driving restriction has different impacts on different pollutants.

3.3 Contribution of this Paper

This thesis will use a different data set from previous literature. To the best of my knowledge, it has not been used for such an assessment previously. It also presents additional evidence by showing the proposed substitution activities of households including reviewing whether households bought additional vehicles, opted for alternative modes of transport or whether there was an intertemporal substitution by households to avoid the restrictions. Contrary to other papers, this thesis assesses the long-term impacts of the policy. Previous papers have made use of RDD or OLS to assess the short or medium-term effect on air quality.

(14)

13

4. Data and Methodology

In this section, I will first introduce my theoretical model and hypotheses on the potential impact of the driving restrictions. I will then summarize the data used in the empirical evidence and the empirical strategy that I used.

4.1 Theoretical Model

This thesis uses a neo-classical household preference model to understand household’s choices of when to drive. This model can show at what point households substitute their driving preferences due to the imposition of a driving restriction like Rodizio Circular.

Households within Sao Paulo will receive private benefits (Bi) and incur private costs12 (Ci)

from driving at certain hours, on certain days of the week. Hence, their private utility will be

the trade-off of these costs and benefits from driving13:

𝑈𝑖(𝐴𝑖) = 𝐵𝑖(𝐴𝑖) − 𝐶𝑖(𝐴𝑖)

Household i drives on a day of the week d at hour t. Each household is restricted from driving on a certain day (d) at an hour t. The accumulation of their driving in a week will be equal to

A (=Aidt). With every additional mile driven by a household from Sao Paulo, there will be

associated damage, like air pollution, equal to D(A). In a perfect information situation, the

policy maker would impose a fine or charge equal to the cost of the marginal damages (𝜕𝐷(𝐴)

𝜕𝐴_𝑖𝑑𝑡) from an individual household i driving on certain days d at certain hours t.

In reality, it is hard for a policy maker to ascertain what such a fine should be. The driving

restriction of Rodizio Circular ( 𝛾_𝑖, 𝜇_𝑖) involves imposing periods of hours 𝜇_𝑖 on certain days

𝛾_𝑖 for certain households where 𝐴_𝑖= 0. Hence, they will not be allowed to drive within the

restricted area at all. Given that households are likely to still get utility from driving at these hours, they are faced by an optimization of utility (across the period of a week) as follows:

12_{Private benefits include getting to a desired location at a certain hour. Private costs include fuel costs, taxes or} the opportunity cost of the time travelling.

13_{The benefits a household receives will be concave (reducing in every additional mile driven). Private costs} will be convex in regard to the driven amount by household i. Hence, U(.) will be concave overall i.e. with every additional mile driven we would expect the marginal utility of the household to be falling.

(15)

14 𝑀𝑎𝑥_𝐴_𝑖 𝑈_𝑖(𝐴_𝑖) = 𝐵_𝑖(𝐴_𝑖) − 𝐶_𝑖(𝐴_𝑖)

𝑠. 𝑡. 𝐴𝑖𝑑𝑡 = 0 ∶ 𝜆𝑖𝑑𝑡 ∀𝑑∈ 𝛾𝑖, ⋀ 𝑡 ∈ 𝜇𝑖

Where 𝐴_𝑖𝑑𝑡 indicates the hours when such a household is restricted from driving and 𝜆𝑖𝑑𝑡 is the

multiplier for the Lagrange function. The first order conditions for the regulated optimum (𝐴_𝑖𝑅)

are as follows: 𝜕𝛽_𝑖(𝐴_𝑖𝑅₎ 𝜕𝐴𝑖𝑑𝑡 = 𝜕𝐶𝑖(𝐴𝑖 𝑅₎ 𝜕𝐴𝑖𝑑𝑡 ∀𝑑∉𝛾𝑖, 𝑡∉𝜇𝑖 𝜕𝛽_𝑖(𝐴_𝑖𝑅₎ 𝜕𝐴𝑖𝑑𝑡 = 𝜕𝐶𝑖(𝐴𝑖 𝑅₎ 𝜕𝐴𝑖𝑑𝑡 + 𝜆𝑖𝑑𝑡 ∀𝑑∈𝛾𝑖, ⋀ 𝑡 ∈ 𝜇𝑖 𝐴_𝑖𝑑𝑡𝑅 = 0 ∀𝑑 ∈𝛾_𝑖, ⋀ 𝑡 ∈ 𝜇_𝑖

From the first order conditions, we derive the substitution effect between driving in restricted (𝑑′∈𝛾_𝑖, 𝑡′∈ 𝜇_𝑖) and unrestricted hours (𝑑 ∉𝛾_𝑖, 𝑡∉𝜇_𝑖).

𝜕 2_𝑢 𝑖(𝐴𝑖) 𝜕𝐴𝑖𝑑𝑡𝜕𝐴𝑖𝑑′_𝑡′ < 0 ∀𝑑 ∈𝛾𝑖, 𝑡∉𝜇𝑖, 𝑑 ′_∈_𝛾 𝑖, 𝑡′∈𝜇𝑖 If the substitution effect is large enough, households could end up increasing their hours of driving. In theory, this could increase the levels of pollution. There are two potential intertemporal substitution effects which can be assessed. First, households substituting driving to unrestricted hours on the same day i.e. not during the morning restriction of 7am to 10am and neither during the evening restriction of 5pm to 8pm. Second, households driving on the weekends when there are no driving restrictions at any hour. I will also look for evidence that households have adjusted behaviour to circumvent the restrictions by buying an additional car (with a different number plate) or another mode of transport that is not included in the restriction (motorcycle or moped).

4.2. Data Set

The data was exported from the online depository of CETESB (Companhia Ambiental de Estado de Sao Paulo). The CETESB is the state government environment protection agency set up in July 1968 who are responsible for the control, monitoring and licensing of pollution generating activities. They have an extensive online library of hourly observations from 21

(16)

15 monitoring stations in the state of Sao Paulo dating back to 1998. However, many of these stations measure only a few pollutants or meteorological variables. Even fewer have any readings dating back before 2000. Only from 2013 onwards were hourly readings at monitoring stations being consistently produced. Therefore, the data used is from those stations that were reading and publishing hourly readings of the three pollutants of focus. Those being, Ozone

(O3), particle matter with a diameter of 2.5mm (Pm2.5) and Nitrogen Oxides (NOx), were

available at 13, 5 and 10 monitoring stations respectively. The data is from the years 2013 to the start of 2017.

Rodizio Circular is aimed at reducing congestion and pollution. Like Gallego et al (2013a), this thesis uses air pollution as a proxy for the congestion. Gallego et al used Carbon Monoxide

(CO) whereas Ozone (O3), Nitrogen Oxides (NOx) and Particle Matter (Pm2.5) are used in this

thesis14. Table I displays the summary statistics for the three pollutants. The number of observations per pollutant vary dependent on how many monitoring stations took readings and

how much missing data there was.15 Referring to Table I, we can see that the variation in the

levels of pollutant concentration is substantial. The maximum values show extreme levels of pollution that Sao Paulo faced across the period with Nitrogen Oxides having the most

volatility in levels (standard deviation of 53.45). Particle matter (Pm2.5) was less volatile but

still managed to reach extremely high levels of 543ug/m3_{. That is over 50 times the WHO} recommended annual mean and more than 20 times the recommended daily allowance of Pm2.5.16

In Table I data was included on the control variables for temperature and solar radiation. As

referenced in Section 2.3, these meteorological variables are key determinants of O3 formation

and hence, hourly readings were used in the core OLS regressions. However, previous papers (Karatzas, Slini & Papadopolous, 2003; Davis, 2008) have been able to include a wind speed variable also, which has an influence on measuring pollutant concentrations. However, this data was unavailable from the CETESB database. This has the potential to bias the results as wind speed plays a role in how static pollutants are within an area. Therefore, if wind speed is greater it will impact the pollutant count by dispersing the pollution across a greater area.

14_{Each pollutant can be used as a proxy for vehicle frequency. The majority of their concentration is attributable} to vehicles in Sao Paulo as explained in section 2.3

15_{Missing data could occur due to a number of causes (e.g. damage to monitoring station). Any missing data} was excluded from OLS regressions and graphs.

(17)

16 Table I: Summary Statistics

In this thesis, there is not a randomly assigned treatment to two identical groups. The existence of such natural experiment is very rare. The complex nature of atmospheric chemistry and the problem of isolating the effects of individual pollutant emission sources mean it is difficult to get a good treatment and control group. It would be very difficult to find a “control city” which had similar characteristics to the treated city. Hence, the treated are those observations during the time when the driving restriction is in place within the area of Rodizio Circular. The control groups are observations for any time outside that restriction (and outside the restriction area). The control group is not statistically similar to the treatment and hence, the p values of mean independence test between the two groups is nil. While there are similarities between the groups, it would be preferential to have more closely characterised groups. To ensure the results are as informative as possible the results look at three contrasting sub- groups within the data set; times when the driving restriction is in place, in and outside the area of Rodizio Circular, and weekdays and the weekend. These distinctive sub-groups within the data form the basis of the suggestive empirical evidence used to evaluate the theoretical model.

(18)

17 Table II: Descriptive Statistics: Treatment & Control Groups

4.3 Empirical Methodology

For this analysis, the empirical strategy is as follows. Firstly, a linear OLS regression with fixed effects model is used to measure the pollution concentration during the hours of the driving restriction. Further to this, the data is used to show the trend of pollution concentrations across the years under observation.

The relationship between pollution concentration and the Rodizio Circular driving restriction could be characterised as follows:

𝑙𝑛𝑥_𝑖𝑗𝑡 = 𝛼₁𝐷_1𝑡+ 𝛼₂𝐷_2𝑡+ 𝜀_𝑖𝑗𝑡 (1)

The dependent variable 𝑥𝑖𝑗𝑡 is the amount of the pollutant i at station j at time t17. The

explanatory variables included are as follows. 𝐷_1𝑡 and 𝐷_2𝑡 are dummy variables that represent

the two driving restrictions in place (7am-10am and 5pm-8pm respectively). However, the

results of this regression would be biased. They would be upwardly biased for O3 as it forms

at a greater rate during the middle of the day. Hence, we require further omitted variables to reduce this bias and the addition of fixed effects for the monitoring stations.

17_{Logarithmic transformation used as it creates a normally distributed dependent variable which is necessary in} linear regression. See appendix for the transformation impact on distribution of dependent variable.

(19)

18 𝑙𝑛𝑥_𝑖𝑗𝑡 = 𝛽₁𝐷_1𝑡+ 𝛽₂𝐷_2𝑡+ 𝛽₃𝐻_𝑡+ 𝜑_𝑖+ 𝜀_𝑖𝑗𝑡 (2)

𝐻_𝑡 is a vector of covariates that includes indicator variables for the year, month and day of the

week as well as meteorological variables such as temperature and solar radiation. The monthly

dummies account for seasonal variation in pollutant concentrations. 𝜑_𝑖 represents the station

fixed effects to control for the expected heterogeneity between stations. Standard errors are clustered around the entity of monitoring stations.

The benefits of using fixed effects is the ability to control for unobserved heterogeneity when it is time invariant and correlated with independent variables (Stock & Watson, 2015). More specific to this thesis, one of the units of observation is the hourly average pollution level by station. Hence, using stations as the entity and controlling for time-invariant station heterogeneity reduces disproportionate changes in reporting levels at stations that could create bias in the results. For example, Santana station is located close to Sao Paulo airport. Aviation travel can contribute significantly to some of the pollutants measured and therefore it is likely that the readings will be higher on average nearer to the airport. Allowing for station fixed effects reduces the potential bias inherent in the location of this monitoring station. This is the same for the time fixed effects. There is time invariant heterogeneity between the level of pollution on different days of the week, months and across years. The indicator variables account for this heterogeneity and reduce inherent bias that would occur if they weren’t included.

The coefficients of interest are 𝛽1and 𝛽2. They are the coefficients of the interaction terms and

measure the average treatment effect (ATE) of the driving restriction. The interaction is between dummy variables for Monday- Friday, the time of day when the driving restriction is in force and a dummy variable for those monitoring stations with Rodizio Circular. These coefficients show the effect of the driving restrictions on the log hourly average reading of the pollutants. If these are negative, it would suggest that during the driving hours, the concentration of pollutants will be much lower and the Rodizio Circular policy is having an effect. However, alternatively, if positive, the contrary will be the case. Nonetheless, conclusions drawn from these regressions come with caveats. We can only define whether during the restricted hours pollutants have fallen. Further graphical evidence is used to demonstrate the hypothesised household behavioural responses in section 4.1. This will include reviewing the impact of average levels of pollutants on days of the week and hours of the day across the various years (2013-17).

(20)

19

5. Results

In this section, the analysis will begin by looking at the OLS results from the linear regression outlined in the Empirical Strategy section (4.3). This fixed-effects model regression takes account of the effect on the pollutant concentration during the hours when the driving restriction is place. That is, on weekdays between the hours of 7am and 10am and 5pm to 8pm within the Rodizio Circular area. The tables show the coefficients of two dummy variables, one for each driving restriction time. In these regressions, the coefficients show the driving

restriction had a negative effect on O3 formation, no effect on Pm2.5 and lead to an increase in

NOx. The following sub-section supplements the regression tables with graphs showing the average levels of different pollutants have trended between 2013 and 2017. A final sub-section presents some tests on the robustness of the regression results.

5.1 Driving Restriction Impact on Pollutant Concentrations (NOx, Pm2.5 and O3)

5.1.1 Ozone Results

Table III below shows the results for ozone concentration during the hours that Rodizio

Circular restrictions are in place. The first row is equivalent to the β1 in equation 2 in section

4.3. The second row represents β2 in the same equation. The dependent variable is the log

(21)

20

Table III: Impact of Rodizio Circular Driving Restrictions on O3 Pollution Levels

Firstly, the effect of the morning driving restriction (7am-10am) is shown in the top row of

Table II. This coefficient indicates that the driving restriction reduced pollution of O3

significantly within these hours. The coefficient is equal to 0.752 (when all control covariates

are included) and can be interpreted as O3 concentration falling by 75%. This is a reasonably

surprising result. Previous literature had found some effect but not reductions of this magnitude. One might have expected the size of this coefficient to fall with the addition of

covariates for temperature and solar radiation (as O3 formation is greatest when these variables

are highest). However, the addition of these covariates does not reduce the value of the estimator substantially (0.011 difference). The evening driving restriction (5pm – 8pm) does not have any significant results until the control variable for solar radiation is included in the regression (column 4). Reviewing columns (4) and (5), we can see that during the restricted hours the sign of the coefficient changes and ozone levels go up. This runs contrary to the effect of the morning restriction and contrary to the intended outcome of the policy. Inferring some potential reasons for these results, it could be easier to avoid the restricted area within the morning hours and substitute to another mode of transport or drive at a different time. This substitution of activity may be less possible in the evening when commuting out of the centre is unavoidable to return home. Alternatively, the regression could be missing key variables

(22)

21 around different emitters. As previously highlighted, vehicle emissions account for 70.8% of VOCs in Sao Paulo. However, the alternative c.30% of emitters could bias the results of this regression if they are not consistent in their emissions across the day. However, the results

suggest that the driving restriction can have a significant impact on reducing O3 (particularly

in the morning restriction). Whereas, the results show an increase during the hours of the evening restriction of c.12%.

5.1.2 Pm2.5 Results

This table shows the effect on Pm2.5 concentrations within Rodizio Circular during the hours

of the driving restrictions. The dependent variable is log hourly average reading of Pm2.5:

Table IV: Impact of Rodizio Circular Driving Restrictions on Pm2.5 Pollution Levels

In first row, all the coefficients are positive which would hypothetically suggest that particle

matter (Pm2.5) increases between 7am and 10am on weekdays. However, none of the estimators

are statistically significant so it would be incorrect to suggest Pm2.5 is increasing during the

hours of the driving restriction. For the evening restriction, the coefficients are also not significant. It would seem that the hour of the day, temperature and solar radiation control

variables are significant18 but all the coefficients are very small. Hence, the table suggests that

18_{Atmospheric studies have shown Pm2.5 to be sensitive to temperature. This is reflected in the results as the} “Temperature” is heavily significant.

(23)

22 the driving restriction of Rodizio Circular has no effect on the count of particle matter (less than 2.5mm).

5.1.3 Nitrogen Oxides Results

Table IV shows that the driving restrictions, in either the morning or the evening, do not reduce the amount of Nitrogen Oxide but rather the pollutant count increases. The coefficient for the morning restriction is statistically significant in all regressions. The addition of further meteorological controls variables does not reduce significance. However, the evening restriction (5pm-8pm) is significant in columns (1) to (3). The level of significance falls with the addition of the “Solar Radiation” variable. The coefficient is insignificant when controls added for monthly and daily of the week variation (column 5).

Table IV: Impact of Rodizio Circular Driving Restrictions on NOx Pollution Levels

5.1.4 What do these results suggest (cumulatively)?

There are no significant results regarding the Pm2.5 concentration during the hours when

Rodizio Circular restrictions are in place. The policy should reduce the number of cars by around 40%. These results suggest this is not the case. Rather, they infer that the Rodizio Circular driving restriction has converse effects on the concentrations of Nitrogen Oxides

(NOx) and Ozone (O3). Examining Table III and O3 results, the concentration is falling during

(24)

23 statistically significant coefficients in Table IV for NOx). These results would support the research of Salvo (2017). He stated that an area can be either VOC-limited or NOx-limited. This atmospheric state, as defined earlier, is key to understanding the ozone formation. Salvo

(2017) found Sao Paulo to be VOC-limited i.e. Ozone (O3) concentrations are increasing in

VOCs and decreasing in Nitrogen Oxides.

Figure 5: Hourly Average Concentration of Pollutants: NOx & O3

1. NOx increasing, O3 falling, VOCs falling.

2. NOx falling, O3 increasing, VOCs increasing (70.8% of VOCs caused by cars in Sao Paulo) N.B. 1 and 2 represent the same time periods in each graph.

As Nitrogen Oxides concentrations go up, O3 goes down as show in the figure above. Figure

5 shows that within Sao Paulo, NOx and O3 concentrations seem to be negatively correlated.

This (alongside the insignificant results for Pm2.5) does suggest that the number of vehicles and

pollution during this observation period is not falling19.Therefore, households must be

circumventing these restrictions. Figure 5 suggests that there is an intertemporal substitution to other times in the day. The subsequent section will look at this and other household behaviour substitutions.

5.2 Pollution 2013-17 in Sao Paulo – Household Behavioural Adjustments

From the tables above, there is evidence that the Rodizio Circular scheme has not had the desired impact of reducing pollution within in Sao Paulo. This section presents some further evidence to support the hypothesis that the policy has been ineffective. In this and the subsequent section, the focus is on how households have been adjusting their behaviour to avoid the restrictions.

19_{This is based on the information that VOCs, NOx and Pm2.5 emissions are primarily from vehicles.}

1

2

1

(25)

24 To understand the trend of pollution levels across the years, the graphs in Figure 6 show the average pollution concentration per hour of the day. The graphs in Figure 6 present how these hourly averages changed, year on year. Examining year on year pollution levels allows understanding of whether the average reading per hour had been falling across the observation period.

(26)

25 In the figure above, the year on year trends of hourly pollution are shown. The bell shape of

the O3 graph reflects that O3 formation is highest when sunlight is greatest (as referenced in

section 2.3). The levels of O3 have increased across the period but have fluctuated upward from

2013 to 2014 before reducing again in 2015 and experienced a further reduction in 2016. Across the period of 2013 to 2017, the levels of Ozone have increased. Table III, in section 5.1.1, displayed that the driving restriction reduced Ozone levels in the hours of restriction. However,

this graph suggests that, overall, the amount of O3 air pollution has been trending upward across

the period. This can be attributed to households substituting driving times to outside the driving restriction period.

If we look at the graph for Ozone (O3). The differential between the yearly averages is greatest

at the peak of the bell shape i.e. when the sunlight is greatest (around the middle of the day). Hence, it would appear that households have increased driving in the afternoon. This behaviour is potentially driven by efforts to avoid the early morning and early evening driving restrictions. Hence, due to the particular nature of ozone formation, Rodizio Circular would appear to be

increasing O3 levels across the period as households substitute their driving to the middle of

the day. This is when the marginal damages of O3 are greatest from the higher frequency of

formation.

For both NOx and Pm2.5 there is an upward trend from 2013 to 2014. The average levels of

pollution then fall, as they did with O3, however, they fall below 2013 levels. Therefore, across

the period of observation, the average levels of PM2.5 and NOx are falling. We could attribute

(27)

26 (18 years after its implementation). Rather, some of this success could be attributed to other policies in Sao Paulo coming to fruition. For example, then Mayor, Fernando Haddad began a huge program in 2014 to make Sao Paulo’s centre (an area inclusive of the area of Rodizio Circular) as vehicle free as possible. He added hundreds of miles of cycle lanes, widened

pavements, added many bus lanes, even closed off roads to traffic and reduced speed limits20.

Litman et al (2011) suggest that such policies have a lag of 1 to 3 years to take effect. Hence, perhaps we are seeing the success of Mayor Haddad rather than a delayed impact of Rodizio Circular.

5.3 Further Evidence of Changes in Household Behaviour.

Another possible time that households could substitute their driving to is the weekend (when Rodizio circular isn’t in effect). Taking Ozone concentrations as an example, Figure 6 presents the average trend across the week for those monitoring stations inside and outside the restricted area.

Shown in the graph below are average daily readings inside and outside Rodizio Circular area.

Areas outside Rodizio Circular have higher daily average levels of O3. Concentrations are

highest on the weekend across Sao Paulo, however, the increase in the pollutant count is greatest within Rodizio Circular. This is suggestive that households are substituting their driving to the weekends within Rodizio Circular to avoid the restrictions.

20_{Berechmann (2009) found that low speed congestion (common in many cities) was a major contributor to} emissions.

(28)

27 Figure 7: Average Level of Pollution by Day of the Week

Furthermore, the suggestion from previous literature in reviewing road rationing schemes is that households have subverted regulations by buying alternate modes of transport (motorbikes) or additional vehicles (Davis 2008, Lawell et al, 2017). Using monthly data on the monthly amount of different private owned vehicle bought from 2008 to 2017, Figure 7 shows the level of car ownership has steadily increased across the period. In the same period,

the population of Sao Paulo has grown by only 6%21 as opposed to the 31% increase in private

car ownership. Therefore, this presents suggestive evidence that households have been buying additional cars to circumvent the restrictions of Rodizio Circular. If they haven’t been buying additional cars then they may have been buying alternative modes of transport, in the form of mopeds or motorcycles. There is again suggestive evidence to support this. In Figure 8 we can see purchases of mopeds or motorcycles have increased by c.70% from 2008 to 2017, far outstripping population growth in the city.

21_{Sao Paulo population figures are available on the website of Instituto Brasileiro de Geografia e Estatistica} (Brazilian Institute of Geography and Statistics)

(29)

28 Figure 8: Population of Private Cars owned in Sao Paulo (2008-2017)

Figure 9: Population of Private Motorcycles/ Mopeds owned in Sao Paulo (2008-2017)

5.4 Robustness Tests

This section presents alternative specifications to ensure the validity of the core regressions. Namely, one specification without “extreme” readings, another with data for the oil price included and finally, the core regressions without the meteorological data (but keeping all other variables).

(30)

29 Firstly, I used a restricted sample without “extreme” readings. The decision of what is deemed extreme is arbitrary. The regression was run with two different thresholds for “extreme”. One

where I removed all readings above 250 ug/cm3 and another where I lowered the threshold to

150ug/cm3. The former being 25 times the recorded limit and the latter being 15 times.22 The

results are shown in Table V. They show that for each of the pollutants neither the sign, nor the significance has changed from the initial regressions.

Secondly, the regressions were re-run for all the pollutants with the additional control variable of gasoline (or oil prices) included. The intuition being that changes in gasoline prices would

impact the utility function of households23 and hence, the amount of driving they do. However,

when gasoline prices were added, there was little impact on the estimates.

Finally, the regression was run without the meteorological variables (“Temperature” and “Solar Radiation”). Similar to the other robustness tests, Table V displays that neither the size nor the

significance was particularly impacted for either NOx nor Pm2.5. However, for O3, we can see

that the size of the coefficient is reduced. This is fairly expected given the impact, outlined

previously, that meteorological factors have on O3 formation.

22_{It could be argued that a threshold of 150 ug/cm}3_{is still too high but there was no official guidance. Anything} over the daily allowance could feasibly be deemed “extreme”. It is an arbitrary measure of my own design. 23_{If the cost of fuel goes up, this will increase the costs of driving for the household. Theoretically, this may} reduce the amount of driving done by households.

(31)

30 Table V: Robustness Tests of Results

5.5 Data Limitations

Given Rodizio Circular was implemented in 1998 officially, it would have been ideal to have data prior and post its implementation. This would allow use of a regression discontinuity design (RDD) to measure the impact of the policy on the pollution concentrations. This was used to good effect to evaluate both Mexico City’s (Davis 2008) and Bogota’s (Lawell et al, 2017) road rationing efforts. However, as aforementioned, the air pollution readings in Sao Paulo were not available prior to when the policy was implemented.

Furthermore, an improved evaluation of the policy would include some variables on congestion i.e. a measurement of the number of the cars on the road. This would improve isolation of the impact of vehicle emissions rather than other potential contributors. No previous studies have been able to attain such data for a sufficiently long period. However, in future years such data may become more available with ever-increasing tracking of vehicles.

Data on wind speed was not available and this will influence the reading of air pollution concentration. Previously, studies of driving restrictions had included wind speed and found it a significant control variable. However, the data was not available from the CETESB online data depository. In relation to assessment of the behavioural changes, it would have been beneficial to assess public transport use. However, there were no available sources providing data on Sao Paulo public transport populations. This could have highlighted some household

(32)

31 substitution to public transport which would likely reduce the levels of pollution (with less private vehicles being used).

In the absence of a “control city” to compare against Sao Paulo as the “treated city” the evaluation of the treatment will have some bias. Even though the treated area is clearly defined, air pollution is not static and will be influenced by wind strength contaminating the “control” area outside the Rodizio Circular area. The area outside will be affected by the driving restriction and hence, it is not truly controlled for the experiment. Gallego et al (2013) used a synthetic policy using a pre-implementation trial period in Mexico City. However, this wasn’t possible for Rodizio Circular and finding an almost identical city to Sao Paulo is very unlikely given the geographical and historical uniqueness of the city. Therefore, we must rely on suggestive evidence from the controls used despite their tendency to bias results.

6. Conclusion & Potential Public Policy Implications

6.1 Findings & Conclusions

Using this data set from Sao Paulo from 2013-2017 and a fixed effects OLS regression model

the results found that O3 levels were reduced under one restriction (in the morning hours) and

slightly increased under another (in the evening hours). Conversely, using the same model for NOx, the results were that the level of this pollutant increased during the morning driving restriction. The Pm2.5 coefficients were positive but statistically insignificant, suggesting no

effect on this pollutant. The combination of the NOx and O3 results seem to indicate a

VOC-limited atmosphere is present in Sao Paulo.

When reviewing the trends across the years in the data set (in Figure 6), those graphical results contributed to the conclusion that Rodizio Circular was not having the desired impact on air

pollution. Rather for O3, it seemed that pollution levels increased across the period 2013- 2017.

Nonetheless, the volatility in results does not seem to suggest Rodizio Circular is the cause. The accumulation of the regressions and graphs supported the conclusion that Rodizio Circular was ineffective in reducing pollution. However, there are obvious caveats around the results which put limits on the conclusion. Vehicles are not the sole emitters of these pollutants so the results are only suggestive of numbers of vehicles but could be the result of other emitters e.g.

production24, aviation. There were data limitations which were outlined in section 5.5.

Furthermore, Rodizio Circular is not the only policy scheme impacting pollution.

(33)

32 In section 5.2, the discussion was around pollutants levels being on a downward trend from 2015 onward and that other factors may be contributing. The scheme undertaken by Mayor Haddad has been extensive since 201425 and are likely to have had an impact. Of course, Rodizio Circular could be a contributor as well but rather evidence would suggest that it has

been badly designed, particularly when aiming to reduce O3. If the ban does not include peak

Ozone formation times, like the middle of the day, then it will create unintentional behavioural adjustments. Households will either change their behaviour to circumvent the restrictions. If this includes driving more in the afternoon then the damages will be greater and pollution levels will rise overall. The O3 graph in section 5.2 is evidence of this behaviour by households. Furthermore, the data on Sao Paulo households shows that households are continuing to buy more cars and other vehicles (not banned in the restricted hours i.e. motorcycles). Households also seem to be substituting driving to the weekend as shown in Figure 5. Hence, the policy is not reducing pollution but rather changing when emissions happen and in some cases, this is when marginal damages are greatest.

6.2 Potential Policy Improvements & Future Research

Initially, Rodizio Circular was viewed positively and did reduce the level of pollution (Jacobi, Kjellen & Segura, 1999). However, there were too many avenues for households to avoid the restrictions. Reviewing the policy in isolation, it could be dramatically improved by making the driving restriction constant across the week. Potentially this could include the weekend but at least then it should come closer to reducing driving numbers during the weekdays. In addition to Rodizio Circular, the new schemes including pedestrianised areas, cycle lanes and widen footpaths could work in conjunction with the policy to improve air quality within Sao Paulo. Air pollution is not a single chemical substance but rather consists of a cocktail of many pollutants who are interdependent and do not react homogeneously to individual policies. Therefore, any successful attempt to improve air quality in Sao Paulo will require a multi-faceted approach due to the heterogeneity of sources and pollutants, similar to efforts currently ongoing in China (Hao & Wang, 2012). As seen from the results of this thesis, any policy program must attempt to reduce household behavioural changes that lead to further polluting behaviour. It would seem one of the most successful programmes has been Low Emission Zones (LEZs) in Europe. This program has achieved welfare gains and improved air quality (Wolff, 2014). Households responded by investing in cleaner technologies. Hence, this policy

(34)

33 encouraged positive household responses. This may become more viable in Latin America as investment in clean technologies increases.26_{Further research will be required to assess the} effectiveness of such policies outside Europe.

26_{In 2012 Brazil introduced its first electric cars. By 2015, there were 2,214 in Sao Paulo alone. According to} research, Brazil’s demand is expected to hit 80,000 units by 2020.

(35)

34

References

1. Abe, K.C. and Miraglia, S.G.E.K., 2016. Health Impact Assessment of Air Pollution in São Paulo, Brazil. International journal of environmental research and public

health, 13(7), p.694.

2. Berechman, J., 2009. Estimation of the full marginal costs of port related truck traffic. Evaluation and program planning, 32(4), pp.390-396.

3. Davis, L.W., 2008. The effect of driving restrictions on air quality in Mexico City. Journal of Political Economy, 116(1), pp.38-81.

4. Eskeland, G.S. and Feyzioglu, T., 1997. Rationing can backfire: the “day without a car” in Mexico City. The World Bank Economic Review, 11(3), pp.383-408.

5. Gallego, F., Montero, J.P. and Salas, C., 2013. The effect of transport policies on car use: Evidence from Latin American cities. Journal of Public Economics, 107, pp.47-62.

6. Gibson, M. and Carnovale, M., 2015. The effects of road pricing on driver behaviour and air pollution. Journal of Urban Economics, 89, pp.62-73.

7. Gryparis, A., Forsberg, B., Katsouyanni, K., Analitis, A., Touloumi, G., Schwartz, J., Samoli, E., Medina, S., Anderson, H.R., Niciu, E.M. and Wichmann, H.E., 2004. Acute effects of ozone on mortality from the “air pollution and health: a European approach” project. American journal of respiratory and critical care medicine, 170(10), pp.1080-1087.

8. Holman, C., Harrison, R. and Querol, X., 2015. Review of the efficacy of low emission zones to improve urban air quality in European cities. Atmospheric Environment, 111, pp.161-169.

9. Jacobi, P., Segura, D.B. and Kjellén, M., 1999. Governmental responses to air pollution: summary of a study of the implementation of Rodízio in São Paulo. Environment and Urbanization, 11(1), pp.79-88.

10. Jacob, Daniel J, 1999 Introduction to Atmospheric Chemistry. Princeton University Press

11. Litman, T., 2011. Transit price elasticities and cross-elasticities. Victoria Transport Policy Institute, Discussion paper (originally published in the Journal of Public, Transportation7 (2004), 35–58).

12. National Research Council, 2008. Energy futures and urban air pollution: challenges

(36)

35

13. Salvo, Y.W., 2017. Ethanol-Blended Gasoline Policy and Ozone Pollution in Sao Paulo.

14. Slini, T., Karatzas, K. and Papadopoulos, A., 2002. Regression analysis and urban air quality forecasting: An application for the city of Athens. Global Nest, 4(2-3), pp.153-162.

15. Souza, D.Z., Vasconcellos, P.C., Lee, H., Aurela, M., Saarnio, K., Teinilä, K. and Hillamo, R., 2014. Composition of Pm2.5 and PM10 collected at urban sites in

Brazil. Aerosol Air Qual. Res, 14, pp.168-176.

16. Stedman, D.H., 2004. Photochemical ozone formation, simplified. Environmental

Chemistry, 1(2), pp.65-66.

17. Stock, J.H. and Watson, M.W., 2015. Introduction to econometrics. 3rd Edition, Pearson Edition

18. “WHO Air quality guidelines for particulate matter, Ozone, Nitrogen Dioxide and

Sulphur Dioxide” WHO (2005)

19. WHO (World Health Organization). 2005. “WHO Air Quality Guidelines Global

Update 2005: Report on a Working Group Meeting, Bonn, Germany, 18–20 October 2005.” World Health Org., Geneva.

20. WHO (World Health Organization). 2016. “Ambient Air Pollution: A Global

Assessment of Exposure and Burden of Disease.” World Health Org., Geneva.

21. World Bank, 1983. “Location Factors in the Decentralization of Industry: A Survey of

Metropolitan Sao Paulo, Brazil”, World Bank, Washington

22. Wang, S. and Hao, J., 2012. Air quality management in China: Issues, challenges, and options. Journal of Environmental Sciences, 24(1), pp.2-13.

23. Wolff, Hendrik (2014): Keep Your Clunker in the Suburb: Low Emission Zones and

Adoption of Green Vehicles, IZA Discussion Papers, No. 8180

24. Wu Y, Wang R J, Zhou Y, Lin B H, Fu L X, He K B et al., 2011. On-road vehicle emission control in Beijing: past, present, and future. Environmental Science & Technology, 45(1): 147–153.

25. Zhang, W., Lawell, C.Y.C.L. and Umanskaya, V.I., 2017. The effects of license plate-based driving restrictions on air quality: Theory and empirical evidence. Journal of

(37)

The effect of Rodizio Circular in Sao Paulo on air quality and household decisions