Rio de Janeiro Metropolitan Region Rafaela Soares Bueno (ENCE/IBGE) - - PDF document

1

Identifying the sociodemographic profile of workers according to home-to-work travel time in Rio de Janeiro Metropolitan Region

Rafaela Soares Bueno (ENCE/IBGE) Luiz Sá Lucas (MC15 Consultants) 28th International Population Conference of the International Union for the Scientific Study of Population (IUSSP) – 29 October to 4 November 2017 ABSTRACT Daily urban mobility is one of the main challenges facing by large urban conglomerates around the world nowadays. Brazilian researches indicate work as main reason of daily displacement. Therefore, the indicator "time of home-work travel" has become quite useful in socioeconomic inequality and social spatial organization studies. The present work aims to identify sociodemographic profile of workers in Rio de Janeiro Metropolitan Region according to their mobility conditions, which will be measured by home-to-work travel time. Descriptive analysis and a Constrained Cluster Analysis Genetic Algorithm were performed based on 2010 Brazilian Demographic Census data. As result, five different groups were found: rich-educated, informal jobs, commuting at the same city, commuting between cities, and recent migrants groups. The last two groups showed the largest home-to-work travel time (over 1 hour). The spatial configuration of urban territory of Rio de Janeiro Metropolitan Region is one of the main reasons that explains this time. Other sociodemographic features also characterize the profiles. Key words: Daily urban mobility; Home-to-work travel time; Rio de Janeiro Metropolitan Region.

2

• 1. INTRODUCTION

Throughout the twentieth century, some social phenomena – such as cities territorial expansion, transportation system technological advances, society transformation (through demographic transition process), and increase of individual transport, among others – have happened in Brazil, changing the types of urban

• displacements. Those reasons have contributed to become daily urban mobility more

complex, since there was a spatial dissociation between home and work. Nowadays, big cities and urban agglomerates in diverse developed and in developing countries in the world face the daily urban mobility as a challenge, since they have passed for the same phases during their growing. In general, the Brazilian metropolis have a good infrastructure and urban equipment in their downtown area, such as transportation services and, for this reason, those parts in the cities are more valued and are also more expensive, being occupied by upper-middle and rich classes. Inversely, there is a dearth of State attention in the peripheral areas and, consequently, services and urban infrastructure are worse than central areas. This results in low price of the distant lands and housing and appropriation by low-income people. However, in big Brazilian cities, a plenty of upper-middle classes and high level of education people have preferred to live in distant areas – in residential condominiums – in search for better quality of life, safety and privacy, but they have also preferred to reside in valued areas of the city. Thus, different profiles of Brazilian population have different mobility conditions. Rich and poor people, white or non-white ones, young or old-aged have spent different home-to-work travel time, due to, for example, distance between the places, transportation mode and road conditions. These reasons have conditioned daily urban mobility and, mainly, well-being of people. Therefore, the main objective of this paper is identify profile of workers in Rio de Janeiro Metropolitan Region (RJMR) more likely to spend shorter and longer time on daily travels between home and work, according to sociodemographic features (such as gender, age, color/ race, occupation, average income, level of education, etc.). For this, 2010 Brazilian Demographic Census data were used to perform the quantitative analyzes. This research is based on hypothesis that the recent migrants in RJMR have worse mobility conditions than non-recent migrants, that is, native people and migrants who live

3

in RJMR over five years. According to Breheny (1999, apud Axisa, Scott, Newbold 2012), recent migrants travel longer distances than residents/ people already established for the longest time, due to period of adaptation that recent migrant requires to settle down himself in the new place. In this research, worse mobility conditions will be measured by home-to-work travel time over 1 hour. Moreover, people who lived in some city of RJMR less than 5 years (2010 Brazilian Demographic Census is the reference) were defined as recent migrants. Beside this introductory section, the paper is divided in other four parts. The next section presents an overview about the link among industrialization, urbanization and daily urban mobility in RJMR, since its origin. In addition, it shows up the daily urban mobility according to some sociodemographic features. The third section briefs source of data and methodology used to reach the objective of the research. The fourth section presents the results. Finally, the conclusions are in the fifth and last part.

• 2. LITERATURE REVIEW

The pace of urbanization in upper-middle-income countries have happened quickly in the second half of the twentieth century (UN 2015), such as the Brazilian urbanization process. As a reflection of this fast urbanization process, in the 1960s, the Brazilian urban population had already exceeded, numerically, the rural population, what might be explained through two facts occurred in Brazil in twentieth century: the urban and demographic transitions (Ojima 2016). The urban transition is characterized by rural exodus, owing to agricultural technological advances established in the rural areas, and due to industrialization, especially in Rio de Janeiro and São Paulo cities (Faria 1991; Baeninger 2012). Industrialization has also attracted people from other Brazilian regions, such as Northeast, to both cities until the end of 1960s, approximately. From 1970 on, however, the migration flow towards Rio de Janeiro and São Paulo have decreased due to implement of federal public policies to develop Northeast region through establishment

• f industries and owing to industrial deconcentrate beyond the Rio-São Paulo axis

(Martine 2015). Those public policies have caused the reconfiguration of the Brazilian industry and economy: they have increased the job opportunities to the population and have avoided people to migrate to the Southeast region. Nevertheless, it is important to

4

mention that, although migration flow in direction to Southeast region had decreased, it is still the main region of Brazil, where there are the biggest and most important cities of the country. The demographic transition, defined, in general, by the decrease of mortality and fertility rates, in different phases, has contributed to increase the Brazilian population

• size. At the first moment, the mortality rates have decreased due to urban improvements,

but the fertility rates have kept high in that time. After, the fertility rates have also

• decreased. Because of this gap between both rates, the Brazilian population size has

increased in, approximately, 3% per year, in 1950s, one of the biggest growth rates in history of the country. Some cities have reached 7% per year of population growth. Over later three decades (1980s), the fertility rates have also fallen, and in the twenty-first century, this rate is considered low (less than the reposition level – 2.1 children/ woman). Thus, the fast Brazilian population growth has caused the urbanization and metropolization process, which have occurred at the same time in Brazil (Brito 2006), since more areas were occupied far away from downtown, also implying in the formation

• f other cities. Transport system technological advances have also contributed to urban

sprawl and to process of formation of metropolis, metropolitan areas, and big urban agglomerations. Miralles-Guasch (2002) affirms that the growing of the cities causes changes in their forms and functions and increases the distance between places (such as home and work) and, consequently, time duration to reach these places. Large urban conglomerates in developed and developing countries face this problem nowadays. According to Ojima (2016), urban growth has, essentially, three characteristics: urban population growth, urban area growth, and daily urban mobility integration between urban areas. The two first features can be denominated as urban sprawl, which occurs from expansion of peripheral areas on, since cities’ downtown are denser (Dematteis 1996). This expansion might occur through occupation of new areas with building in real estate expansion zone, and by the incorporation of urban cores (Ojima 2016). As an example, in the peripheral areas, it has been common to note residential condominiums, which contributes to build and settle down centers with several kinds of services, such as malls, education institutes, and business centers. However, these real estate developments

5

generate to another problem related to the transport, since where does not exist public transportation or the same is deficit, the private transportation is stimulated to be used (Limonad 2007). Transport system technological advances provide the connecting of mobility between urban areas and intensify the urbanization process. The commuting between cities has become a common movement for workers and students. In fact, the commuting between cities works as a mechanism that avoid the change of permanent residence (Ojima 2016) for different reasons, such as economy, belonging to somewhere, preference of type of residence, or individual/ familiar choices. For Limonad (2007), the commuting between cities is a survival strategy for different social groups, as well as a new lifestyle for rich people. According to Miralles-Guasch (2002), daily urban mobility patterns are result of individual displacements for diverse goals: work, study, shopping, sports, leisure, health care, etc. Several Brazilian researches indicate that work is the main reason of daily urban

• displacements. For Rodrigues (2015), some demographic features, such as gender, age,

color/ race, level of education, average income, and others condition daily urban mobility. That is, different social groups spend different travel times between residence and work, for instance. Some authors argue that women tend to search for jobs near her residences, due to her domestic responsibilities, children, and vulnerability to violence (Galster; Killen 1995; Flores 2006; Miranda; Rodrigues 2010). Consequently, they tend to spend less travel time between home and work. On the other side, Pereira and Schwanen (2013) have verified, in a study about commuting in the main Brazilian metropolitan regions, that average home-to-work travel time of women and men has been converging nowadays. This fact might be explained by sociodemographic changes in Brazil, such as increase of women’s level of education, inserting women into the labor market, changing in the family lifecycle, increase in the number of households headed by women, and decrease of fertility rate. As indicated by Cunha (2011), the migration relates to vulnerability and socio- spatial segregation because of the loss of set of assets. When people move to another city and decide to settle down in cheaper and peripheral areas in reason for economy, this might be result in to get worse the living conditions, since the sets of assets in these areas

6

have lower quality than the center areas. In this case, set of assets is composed by urban infrastructure, equipment and services, such as buses lines, subway, safety, etc. Moser (1999) defines assets as types of capitals (human, social, and physic) generated by the State, society or market, aiming to auxiliary to increase the welfare level or to reduce the vulnerability. According to Cavalieri (1998), several reasons explain the difference in wages among people, such as qualification and years of study, since the increase of level of education is associated with the increase of income. Thus, high income and educated people might lead them to occupy more valued areas of the urban space – with better services and equipment – or to live away on the peripheral areas – in private residential condominiums in search for space, safety, and quality of life (Ojima et al. 2010). For Miranda and Rodrigues (2010), currently, there is no linear relationship between travel time and income, since families have different preferences from each

• ther. But Pereira and Schwanen (2013) verified that poorest families that live

predominantly in distant areas (due to the low prices of lands and houses) spend, on average, 20% more traffic time than the richest segment, since they are more vulnerable to many problems, such as public transportation. The features related to occupation and type of displacement will be described in the context of RJMR for being particular characteristics of the region. As reported by Pero and Mihessen (2012), 74% of the dwellers of Rio de Janeiro state live in RJMR, and 55% of them work in Rio de Janeiro city. In addition, according to Medeiros Jr. and Grand

• Jr. (2011), in 2008, there were over 460,000 formal jobs of the labor market (employee

with a formal contract and militaries/ statutory civil servants) in Rio de Janeiro city’s

• downtown. The large concentration of formal jobs in this part of the city demonstrates

the intense flow towards there, which overloading the transportation system. The authors Pero and Mihessen (2012) also affirm that, among people who

• ccupied formal jobs on the labor market in eight Brazilian metropolitan regions in 2010,

the percentage of people spending over 1 hour from home to work was higher in RJMR than other regions, especially by people who lived in peripheral areas.

7

Rio de Janeiro Metropolitan Region Rio de Janeiro Metropolitan Region, composed, nowadays, by nineteen cities, was founded about 1920 and its initial point was Rio de Janeiro city, then the capital of Brazil. Two types of transport (trams and trains) were responsible for have transformed the urban space of the city, and have given to it a spatial stratified structure according to social classes (Abreu 2013). Trams have moved around rich areas, while trains have gone to distant areas characterized as rural areas. Throughout the twentieth century, architectonical and urbanistic developments were executed in Rio de Janeiro city, most of which in a specific region: the South Zone. Between 1902 and 1906, the South Zone received a lot of urban improvements, such as

• pening of streets, avenues, and tunnels; construction of important buildings; removal of

urban garbage; sewer services; urban afforestation; improvement of squares by gardening, etc. Otherwise, the urban progress arrived in North Zone by paving of streets, in order to help the employers to carry their cargo to the central harbor. For the population itself there were no improvements (Abreu 2013). The following municipalities’ governments have continued to invest efforts in South Zone through infrastructure improvement and construction of avenues, tunnels, and viaducts, since automotive vehicle flow increased from there to downtown, which was – and is still – the labor and financial center of the city. The expansion of Rio de Janeiro city beyond the South Zone has resulted in formation of the West Zone. Avenues and tunnels were constructed to link these zones of Rio de Janeiro city. This segregated structure remains the same, currently. The South Zone of Rio de Janeiro city is the most expensive area of all Rio de Janeiro Metropolitan Region. This area receives Estate investments in infrastructure, urban equipment and services, and for this reason, it is more valued and thus is occupied by upper-middle, rich and wealthy

• people. Beside investments, other attributes also contribute to give such importance to

some neighbors or regions in the cities, such as natural beauty, privilege, and social status. The high prices of land and realty leave poor people out the segregated area. The “Circular Causality Theory” explains this mechanism of inclusion and exclusion of part of the population in some segregated area (Ribeiro 2016). However, in Rio de Janeiro city it is common to note, due to the distribution of hills through the city, slums all around it, even in South Zone. Therefore, rich and wealthy

8

people coexist in the same area with dwellers of the slums. Both groups might use the same urban equipment and spaces, such as public transportation system, squares, etc. Because of this fact, Ribeiro (2016) argues that hypothesis based on Circular Causality Theory are difficult to apply in Rio de Janeiro city. Through a periphery-center perspective (macroscale), the hypothesis is, perfectly, identifiable. However, the same is not clearly seen on a microscale optical, due to the slums in the rich neighborhoods. Rio de Janeiro Metropolitan Region is the second most populous urban agglomerate in Brazil. Although subway and train lines, boat services and intercity buses compose the public transportation system in RJMR, it is not efficient for all users, especially for those living in distant areas.

• 3. DATA AND METHODOLOGY

In the machine learning, pattern analysis, data science, analytics, and data mining area techniques can be organized into four distinct family of methods: supervised, unsupervised, semi-supervised, and data visualization tools. In supervised learning methods, we have, for each observation in data, one or more input variables and a response variable (both variables from nominal to ratio scale). These kind of tools are also called predictive models. When, for any reason, some instances of these response variables are unknown, these cases are called unlabeled. The objective is to fit an algorithm, based on a statistical model or model-free, so that after fitting the algorithm on a training-test set, where all cases are labeled, we can estimate the response for new data, where we know only the input variables. Linear, multinomial, neural networks and other techniques belong to this class (Witten; Frank 2005; Torgo 2017; Lanz 2015). In unsupervised tools, we do not have a response variable. Based only on a set of input variables, we look for groups or clusters of observations so that they have some characteristic in common (small distances or bigger similarity, most commonly). Here the most used techniques are clustering algorithms (Kaufman; Rousseeuw 2005; Wedel; Kamakura 2000; Venables; Ripley 2002). Semi-supervised tools are situated in a middle-way between the two techniques

• above. We can have here, for instance, a data set where some observations are labeled

and some are not, and we want to label the unlabeled ones. One common technique here

9

is self-training, where we train a supervised tool with the labeled observations, estimate and associate the estimated labels for observations where the accuracy level is good enough, and iterate the process, labeling unlabeled observations, until a stopping criterion is satisfied. Another technique, the one used here, is constrained clustering where we usually constrain pairs of observations to belong to the same group. (Chapelle; Scholkopf; Zien 2006; Liu et al 2006; Zhu; Wang; Li 2010; Yi et al. 2015; Torgo 2017; Vallejo- Huanga; Morillo; Ferri 2017). Data Visualization tools is another set of tools that is growing in use nowadays. Here we have techniques that allow grasping visually the interrelations of quite complex

• data. We will have some examples of these graphs in this study, such as heatmaps, mosaic

plots, and parallel coordinate plots (Kirk 2012; Yau 2013; Kirk 2016; Camões 2016). The work described here is essentially the application of an unsupervised learning clustering method to a sample of the 2010 Brazilian Demographic Census (censuses in Brazil are carried each 10 years) with the objective to analyze the urban mobility of the Rio de Janeiro Metropolitan Region citizens, based on the theory exposed before. Therefore, we have, in a way, home-to-work travel time as the response variable, against explanatory variables: gender, age, color/ race, origin (recent migrant or not), level of education, average monthly income, position in the occupation, type of daily urban mobility, and time spent in home-to-work travel. This work considers people from 16 to 69 years old, occupied during the reference week of the census, who worked outside home and returned home daily, regardless of the fact that the workplace was in the same city where they lived or not, in all cities of RJMR. This screening is necessary to consider the same urban mobility system used by the population. Our semi-supervised approach consisted on, essentially, to apply a cluster analysis technique to data, but in a way that the resulting clusters are ordered by travel time. The idea is not new, and has been applied, for example, in Kamakura and Mazon (2013). To achieve that we based ourselves on the cluster analysis problem as described by the p-median problem (Daskin 2013). To this problem, we added the constraint that the group averages should increase as we go from one group to the other. The p-median problem is known to be NP-complete (see Daskin 2013), what means that, even for small problems we are not aware of any exact solution that can converge to a solution in a

10

reasonable amount of time. NP-complete means that the time for convergence is Non

• Polynomial. Given the computational demand of the problem, we relied on a Genetic

Algorithm (GA), where to a GA algorithm created by us we add the penalty corresponding to the ordering of the groups travel time means. Here we implemented the algorithm using the “GA” package from R software (Scrucca 2013; Scrucca 2016). On the other side, we believe that it does not make sense to only segment a given data base, without developing a classifier, a predictive model, or supervised learning tool, to classify new data into the first segmentation. As there is always an “error” in the predictive model (a difference between the original segments and those given by the classifier - usually the solutions agree on around 70% of the cases), we use the classifier as the “real” segmenter. The cluster analysis solution is only a step-0 in the whole algorithm. As the classifier, we used the Random Forest predictive model, as implemented in the R software package RandomForest (Liaw; Wiener 2002). One interesting feature in this method is that it is able to provide a measure of the input variable importance in the classification. We adopted as the final solution a five segments one (92.0% of hit rate). For the definition of the GA step-0 solution, we had a sample of 1,000 observations. The final classified Census data sample had almost 222,000 observations.

Table 1: Hit Rate Statistic and importance of the variables Number of Groups Hit Rate Importance of the variables Gender Age Race Recent Migrants Level of Education Average income Position in Occupation Type of Daily Urban Mobility Home-to-work travel time 5 92.0 18.9 2.5 1.6 8.9 12.9 4.7 19.6 24.9 6.1 6 94.0 3.0 2.6 0.9

• 1.4

0.3 87.9 1.0 3.7 1.9 7 93.3 24.5 4.2 17.8 4.9 23.5 1.7 0.5 7.9 14.9 8 91.0 22.9 1.9 21.8 3.4 20.6 1.5 0.9 4.8 22.2 Source: IBGE (2010). Microdata of Sample – 2010 Brazilian Demographic Census

As it might be observed on the Table 2 and on the Graph 1, the most important variables of the five-group classification were mobility (24.9%), position in occupation

11

(19.6%), and gender (18.9%), followed by education (12.9%) and recent migrant condition (8.9%).

Graph 1: Importance of the input variables (%) according to classification of 5 groups

The graphs below show, respectively, the difference in the groups means in the 222K sample and the distribution (%) of elements in each group. Graph 2: Means of each group Graph 3: Percentage distribution of elements in each group

Groups %

12

As shown the Graph 3, the groups have different size in number of persons. Group 1 has over 40% of the sample, whereas Group 3 has about 5%. In what follow, the groups will be characterized according to the sociodemographic features and will receive their labels.

• 4. RESULTS AND DISCUSSION

The universe of analysis of this research is composed mainly of men (57.72%), young up to 30 years old (32.86%), white people (46.22%), non-recent migrants (92.94%), people with completed High School (38.29%), occupants of formal positions into labor market (95.98%), average monthly income between 1 and 2 minimum wages (MW) (52.32%). Almost 78.00% lived and worked in the same city and 33.41% spent

• ver half an hour to one hour to move from home to work, according to Table 2 (Annex).

In order to describe each of the groups based on sociodemographic variables, a data visualization technique was used to complement the quantitative analysis: a mosaic plot, which was produced with the aid of the R package Extracat (Pilhoefer; Unwin 2013). Figure 1 shows the sizes of the groups according to the variables and their respective factors. The red and blue colors correspond, respectively, to below-average and above-average ratios, based on statistical hypothesis test. The five groups were labeled as:  IM: Intramunicipal commuting  RE: Rich-Educated people  IN: Informal jobs  MI: Recent migrants  PNPP: Peripheral-nucleus and Peripheral-peripheral commuting In an initial analysis of such graphs, we can see that some groups are quantitatively larger than others, that is, some groups have a large number of people, such as the IM, RE and PNPP groups, as already shown in Graph 3. The names of the groups were chosen based on their predominant variables. Thus, the IM Group has as main characteristic the fact that the worker lived and worked at the same municipality, whatever the municipality of the RJMR, performing, then, intramunicipal movement. The IN Group is mainly composed of people who occupied informal jobs in the labor market, such as employers, employees without a formal

13

contract, and self-employed workers. The MI Group is the group of recent migrants, that is, people living in a municipality of the RJMR for less than 5 years (using the 2010 Demographic Census search date as reference). The PNPP Group is characterized for the main urban daily movements carried out by the people: periphery-capital and periphery- periphery, while the RE Group is made up by highest level of education and purchasing power people among the five groups presented. After a preliminary graphical analysis of the groups, a more detailed description

• f each of them will be performed with the aid of new graphs, which are denominated

Figure 1: Population size of each group

14

parallel coordinated plots. These graphs, besides providing the increasing importance of the variables in each of the groups (from left to right), indicate the most frequent factors and, thus, a sociodemographic profile is constructed. Those plots were also produced through the R package Extracat (Pilhoefer; Unwin 2013). As show Graph 4, the Intramunicipal Group (IM) has as main characteristics the daily urban mobility type, average monthly income, level of education, sex, and migration condition (in decreasing order of importance). In addition to being made up of people who performed daily intra-municipal commuting, they had a monthly average income between 1 and 2 MW (in 2010, the minimum wage corresponded to US$160), low level

• f education (only 3.61% had complete college), were men and non-recent migrants.

Occupation, age and color/ race variables are less important in this profile. Spending 6 to 30 minutes was the main home-to-work travel time (40.35%), probably by people who lived in small cities. Otherwise, 34.25% of IM group’s people spent between half an hour to one hour, probably by people who lived in Rio de Janeiro city, the biggest city of RJMR. The Peripheral-Nucleus and Peripheral-Peripheral Group (PNPP), on the other hand, has as main characteristics the home-to-work travel time between 1 hour and 2 hours, non-recent migrants, average monthly income between 1 and 2 MW, complete High School, and formal job position workers, as show Graph 5. This group has the highest percentage of home-to-work travel time over 1 hour (65.59% against 31.17% of Graph 4: Parallel Coordinated Plot of Intramunicipal Group

15

the maximum in Migrants Group). Excessive time spent in a single home-to-work way impacts people's well-being and compromises leisure time, study time, social interaction,

• r even household responsibilities. This also works as a social exclusion mechanism since

the people has less time to improve and qualify themselves. Distance between cities, high concentration of formal jobs in Rio de Janeiro city, and excessive amount of vehicles

• verload RJMR's transportation system, causing traffic jam and increasing travel time.

In contrast to the previous groups, the Rich-Educated Group (RE) is highly educated (92.63% had at least a complete High School compared to a mean of 48.26% of the other groups) and had a high purchasing power. In addition, they were women, lived and worked at the same city, non-recent migrants, and occupied formal job positions. The variables age and travel time are not very important for this group and their factors are well distributed, that is, there is no predominant age group (Graph 6). Although the intramunicipal movement encompassed 89.51% of the people, the periphery-nucleus movement presents a relative frequency of 6.41%. It is a trend of the large urban agglomerations to expand territorially and the population occupy areas distant from the

• center. According to Lago and Mammarella (2010), the middle class in Rio de Janeiro

moved between 1991 and 2000 "towards the new frontiers of expansion of the real estate market in the peripheral areas", but the high income people moved to areas most valued in the city. Graph 5: Parallel Coordinated Plot of PNPP Group

16

In this research, it was found a group made up only of recent migrants, thus referred as the Migrant Group. Its main characteristics are: recent migrants who worked in formal positions in the labor market, that is, as employees with a formal contract and military / public civil servants. It is the second group with the highest purchasing power (29.37% had an average monthly income over 3 MW) and education (57.28% had completed High School). Intramunicipal mobility is the main movement, but periphery- nucleus mobility stands out as the second largest type among the five groups (22.71%). The other variables – age, color / race, and gender – do not stand out, and their factors are distributed in a similar way. Nevertheless, looking the numbers (Table 3), the Migrant Group is the second group with the highest percentage of travel time over 1 hour. According to Cunha (2011), spatial mobility is related to vulnerability in the form of loss

• f set of assets, that is, migration to peripheral areas may result in to get worse living

conditions, since in such areas the assets, such as infrastructure, equipment and urban services may be lower quality when compared to the central areas. Graph 6: Parallel Coordinated Plot of Rich-Educated Group

17

Finally, the Informal Group consists, essentially, of people who occupied informal job positions, such as employees without a formal contract, employers, and self-

• employed. Color/ race and average monthly income are the most important features and

the other characteristics do not stand out analytically, which may indicate that people of all genders, ages, level of education, and migration conditions are part of this group. For instance, 25.95% had only incomplete Elementary School, while 55.86% had complete, at least, High School. The intramunicipal movement is also marked in this group (78.62%) and another 12.27% of the people performed peripheral-nucleus displacements. About 22.00% spent between 1 hour and 2 hours in home-to-work travels, but 33.52% spent between 6 and 30 minutes. (Graph 8). Graph 7: Parallel Coordinated Plot of Migrant Group Graph 8: Parallel Coordinated Plot of Informal Group

18

• 5. CONCLUSIONS

The new types of migration flows and population displacements of the twenty- first century are mostly urban-urban and differ from population movements in the second half of the twentieth century. The main distinction is due to the decrease in the intensity

• f the long distance flows, giving rise to short movements, such as the commuting
• displacements. This transformation, accompanied by the political, economic and social

changes of the country, has given complexity to the population movements, since the different population groups and flows have particular characteristics in related to mobility, expressing a peculiar pattern on the displacements. The spatial configuration of the large urban agglomerations themselves causes the complexity of daily population movements. Urban sprawl across the territory, transportation system technological advances, and easy access to the purchase of private vehicles intensify daily commutes. Some people prefer to live in the central areas due to good infrastructure, services and urban equipment provided by the State. Other groups already opt for the peripheral areas, in search for space, privacy and quality of life. There are still people who, involuntarily, have to live in peripheral areas that lack the attention

• f the state - with urban infrastructure of lower quality than the central areas.

The present study identified five population groups, different from each other in terms of daily urban mobility. A construct measured by the variable "usual time spent on home-to-work travel" from the Sample Questionnaire of 2010 Brazilian Demographic Census with other sociodemographic variables, such as gender, age, income, occupation, level of education and others, it was possible to create profiles of workers through a clustering technique using Genetic Algorithm. One of the main variables of the study – type of daily urban mobility – generated two distinct groups: the Intramunicipal group – who lived and worked at the same city – and the group of people who performed commuting between cities and resided in the periphery (Periphery-Nucleus and Periphery-Periphery Group). Both groups have different characteristics from each other. The second group, for instance, recorded high percentages of travel time over 1 hour (65.59% versus 17.12% of the first one). On the

• ther hand, it had higher level of education and purchasing power than the Intramunicipal

Group.

19

Another group identified was the group of the most educated and highest income people than other groups, denominated here of Rich-Educated Group. They spent relatively little time to move from home to work, registering low percentages of time over 1 hour. The Migrant group found supports the research hypothesis that recent migrants have worse daily mobility conditions than the non-recent migrants. The group recorded the second highest percentage of travel time over 1 hour and of periphery-nucleus

• displacement. Vulnerability and loss of assets may be related to such conditions, since

the high cost of Rio de Janeiro city may force recent migrants to settle in outlying cities and use commuting between cities as a survival strategy in the Rio de Janeiro Metropolitan Region. The group of occupants of informal jobs does not have peculiar sociodemographic characteristics, involving all the population groups as to the sex, age, income, etc. The social organization of the RJMR territory, with a high concentration of economic activities in the capital and segregated areas, leads to different conditions of daily urban mobility of the population. As shown in this study, some population groups have worse mobility conditions than others related to home-to-work travel time, and in practice, it is possible to see such conditions in Rio de Janeiro Metropolitan Region. Considering also the macro-scalar and micro-scalar perspectives, that is, center-periphery and intraregional duality, respectively, – it becomes difficult to establish a single pattern capable of explaining the population movement of the Metropolitan Region.

• 6. REFERENCES

Abreu, M. 2013 A. Evolução Urbana do Rio de Janeiro. 4 ed. Rio de Janeiro: Prefeitura da Cidade do Rio de Janeiro/SMU/IPP. Baeninger, R. 2012. Rotatividade Migratória: um novo olhar para as migrações no século

• XXI. Revista Interdisciplinar da Mobilidade Humana, n. 39: 77-100.

Camões, J. 2016. Data at Work: Best practices for creating effective charts and information graphics in Microsoft Excel, San Francisco Pearon Education – Peachpit. Cunha, J.M.P. 2011. Mobilidade espacial, vulnerabilidade e segregação espacial: reflexões a partir do estudo da RM de Campinas, 2007, in: Cunha, J. P (Org). Mobilidade

20

espacial da população: desafios teóricos e metodológicos para o seu estudo. Campinas: Núcleo de Estudos de População – Nepo/Unicamp. Dematteis, G. 1996. Suburbanización y periurbanización: ciudades anglosajonas y ciudades latinas. 9 p. Trabalho apresentado no Seminário La Ciudad Dispersa, Suburbanización y Nuevas Periferias, realizado em Barcelona, 1996. Disponível em: http://archivouel.tripod.com/dispersa.pdf . Accessed on 10 Aug 2017. Demiriz, A., Bennet, M. and Embrechts, M. 1999. Semi-supervised clustering using genetic algorithms, Artificial Neural Networks in Engineering (ANNIE-99). 809-814. New York: ASME Press. Faria, V. 1991. Cinqüenta anos de urbanização no Brasil: tendências e perspectivas. Novos Estudos CEBRAP, São Paulo. Flores, Carolina. 2006. Consequências da Segregação Residencial: teoria e métodos. In: Cunha, J.M.P (Org.). Novas metrópoles paulistas: população, vulnerabilidade e segregação. Campinas: Núcleo de Estudos de População – Nepo/Unicamp. Galster, G.C. and Killen, S.P. 1995. The geography of metropolitan opportunity: a reconnaissance and conceptual framework. Housing police debate, v. 6, n. 1. Hoppner, F. and Klawonn, F. 2008. Clustering with Size Constraints. 137.10.1007/978- 3-540-79474-5_8. IBGE – Instituto Brasileiro de Geografia e Estatística. 2016. Metodologia do Censo Demográfico 2010. Série Relatórios Metodológicos, vol. 41, 2ª ed. Rio de Janeiro. ______. Microdados do Censo Demográfico 2010. 2011. Rio de Janeiro, IBGE. Kamakura, W. and Mazon, J. 2013. Estratificação Socioeconômica e Consumo no Brasil. São Paulo: Editora Relativa. In Portuguese. Kaufman, L. and Rousseuw, P., 2005. Finding Groups in Data: An Introduction to Cluster Analysis. New Jersey: John Wiley & Sons. Kirk, A. 2012. Data Visualization: a successful design practice. Birmingham: Packt Publishing. ______. 2016. Data Visualization: a Handbook for Data Driven Design. London: Sage Publications. Lago, L.C. and Mammarella, R. 2010. Da hierarquia de classes à organização social do espaço intraurbano: um olhar comparativo sobre as grandes metrópoles brasileiras. Cadernos Metrópole, v. 12: 65-84.

21

Lantz, B. 2015. Machine learning with R. Birmingham: Packt Publishing. Liaw, A. and Wiener, M. 2002. Classification and Regression by randomForest. R News 2(3), 18-22. Lima, E.E.C. and Braga, F.G. 2013. Da rotatividade migratória à baixa migração: uma análise dos padrões da mobilidade populacional no Brasil de 1995-2000. Revista Brasileira de Estudos Populacionais, Rio de Janeiro, vol. 30: 55-75. Limonad, E. 2007. Urbanização dispersa: mais uma forma de expressão urbana? Revista Formação, n. 14, vol. 1: 31-45 Liu, J., Zhang Q., Wang, W., McMillan, L., and Prins, J. 2006. Clustering Pair-wise Dissimilarity Data into Partially Ordered Sets. KDD’06, Philadelphia, Pennsylvania, USA. Martine, G. 2015. Prefácio. In: Ojima, R. and Fusco, W. (Org). Migrações nordestinas no século XXI: um panorama recente. Blucher, São Paulo. Medeiros Jr, H. and Grand Jr, J. 2011. Distribuição dos empregos formais na cidade do Rio de Janeiro: uma análise exploratória. Coleção Estudos Cariocas, Instituto Municipal de Urbanismo Pereira Passos, Rio de Janeiro. Miralles-Guasch, C. 2002. Ciudad y transporte, el binomio imperfecto. Ariel Geografia, Barcelona. Moura, R., Castello Branco, M.L.G., Firkowski, O.L.C.F. 2005. Movimento pendular e perspectivas de pesquisas em aglomerados urbanos. São Paulo em perspectiva, v. 19: 121-133. Ojima, R. 2016. Pessoas, prédios e ruas: por uma perspectiva demográfica dos processos urbanos contemporâneos. In: Dispersão urbana e mobilidade populacional. Editora Edgard Blücher Ltda, São Paulo. Pereira, R. and Schwanen, T. 2013. Tempo de deslocamento casa-trabalho no Brasil (1992-2009): diferenças entre regiões metropolitanas, níveis de renda e sexo. Textos para discussão. Brasília, Instituto de Pesquisa Econômica Aplicada. Pilhoefer, A. and Unwin, A. 2013. New Approaches in Visualization of Categorical Data: R Package extracat. Journal

• f

Statistical Software, 53(7), 1-25. http://www.jstatsoft.org/v53/i07/. R Core Team 2017. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.

22

Ribeiro, L.C.Q. 2016. Metamorfoses da ordem urbana da metrópole brasileira: o caso do Rio de Janeiro. Sociologias, Porto Alegre, ano 18, n. 42, pp. 120-160. Rodrigues, J.M. 2015. Condições de mobilidade urbana e organização social do território. In: RIBEIRO, L.C.Q. (Org.). Rio de Janeiro: transformações na ordem urbana. 1 ed. Rio de Janeiro: Letra Capital; Observatório das Metrópoles. Scrucca, L. 2013. GA: A Package for Genetic Algorithms in R. Journal of Statistical Software, 53(4), 1-37. ______. 2016. On some extensions to GA package: hybrid optimization, parallelization and islands evolution. http://arxiv.org/abs/1605.01931. Acessed 22 Aug 2017 Torgo, L. 2017. Data Mining with R: Learning with Case Studies, Boca Raton: CRC Press. UNITED NATIONS, Department of Economic and Social Affairs, Population Division

• 2015. World Urbanization Prospects: The 2014 Revision, (ST/ESA/SER.A/366).

Vallejo-Huanga, D.; Morillo, P.; and Ferri, C. 2017. Semi-Supervised Clustering Algorithms for Grouping Scientific Articles. Procedia Computer Science 108C 325–334. Venables, W. and Ripley, B. 2002. Modern Applied Statistics with S (2017). Data Mining with R: Learning with Case Studies, New York: Springer-Verlag. Villaça, F. 2001. Espaço intra-urbano no Brasil. São Paulo: Studio Nobel: FAPESP: Lincoln Institute. Wedel, M. and Kamakura, W. 2000. Market segmentation: Conceptual and methodological foundations, Dordrecht: Kluwer Publications. Witten, I. and Frank, E. 2005. Data Mining: Practical Machine Learning Toools and

• Techniques. San Francisco: Morgan Kaufmann Publishers.

Yau, N. 2013. Data points: Visualization that Means Something, New Jersey: John Wiley & Sons. Yi, J., Zhang, L.; Tianbao Yang, T.; Liu, W.; Wang, J. 2015. An Efficient Semi-Supervised Clustering Algorithm with Sequential Constraints. KDD’15, Sydney. Zhu, W.; Wang, D.; Li, T. 2010. Data clustering with size constraints, Knowledge-Based Systems, Volume 23, Issue 8, December 2010, Pages 883-888.

23

• 7. ANNEX

Table 2: Distribution of occupied people from 16 to 69 years old who worked outside home and returned to home daily according to sociodemographic variables in the Rio de Janeiro Metropolitan Region, 2010 Variables Factors Number

• f person

% Standard error Gender Male 2,182,728 57.72% 0,0011 Female 1,598,617 42.28% 0,0011 Age Over 16 to 30 years 1,242,587 32.86% 0,0010 Over 31 to 40 years 1,014,649 26.83% 0,0010 Over 41 to 50 years 848,273 22.43% 0,0009 Over 51 to 60 years 527,288 13.94% 0,0008 Over 61 to 69 years 148,548 3.93% 0,0004 Color/ Race White 1,747,861 46.22% 0,0011 Brown 497,945 13.17% 0,0007 Black 1,500,952 39.69% 0,0011 Others 34,587 0.91% 0,0002 Condition

• f migrant

Non recent migrants 3,514,225 92.94% 0,0006 Recent migrants 267120 7.06% 0,0006 Level of Education No instruction/ Incomplete Elementary School 951,072 25.15% 0,0010 Complete Elementary School / Incomplete High School 706,634 18.69% 0,0009 Complete High School / Incomplete College 1,447,734 38.29% 0,0011 Complete College 675,904 17.87% 0,0009 Position of Occupation Formal 3,629,366 95.98% 0,0004 Informal 151,979 4.02% 0,0004 Average monthly income Less than 1 MW 327,103 8.65% 0,0006 Over 1 to 2 MW 1,978,318 52.32% 0,0011 Over 2 to 3 MW 552,794 14.62% 0,0008 Over 3 to 5 MW 395,957 10.47% 0,0007 Over 5 to 10 MW 332,788 8.80% 0,0006 Over 10 MW 194,385 5.14% 0,0005 Type of Daily Urban Mobility Live and work at the same city 2,941,853 77.80% 0,0009 Live at nucleus and work at peripheral city 41,602 1.10% 0,0002 Live at peripheral city and work at nucleus 535,179 14.15% 0,0007 Live and work at different peripheral cities 262,711 6.95% 0,0005 Home-to- work travel time Less than 5 minutes 220,881 5.84% 0,0005 Over 6 to 30 minutes 1,217,537 32.20% 0,0010 Over 31 minutes to 1 hour 1,263,328 33.41% 0,0010 Over 1 hour to 2 hours 874,493 23.13% 0,0009 Over 2 hours 205,106 5.42% 0,0005 Source: IBGE (2010). Microdata of Sample - 2010 Brazilian Demographic Census

24

Table 3: Characteristics of the five groups according to sociodemographic variables in the Rio de Janeiro Metropolitan Region, 2010 Variables Factors IM RE IN MI PNPP Gender Male 73.72% 32.57% 45.66% 61.83% 65.74% Female 26.28% 67.43% 54.34% 38.17% 34.26% Age Over 16 to 30 years 34.58% 24.65% 28.98% 46.39% 39.29% Over 31 to 40 years 25.93% 27.39% 23.62% 28.86% 28.01% Over 41 to 50 years 21.53% 26.01% 23.49% 15.63% 20.56% Over 51 to 60 years 13.66% 17.36% 17.68% 7.39% 10.06% Over 61 to 69 years 4.30% 4.60% 6.22% 1.73% 2.09% Color/ Race White 39.45% 64.20% 48.70% 48.16% 28.20% Brown 16.04% 8.68% 12.09% 11.18% 15.42% Black 43.68% 26.35% 38.06% 39.30% 55.22% Others 0.83% 0.77% 1.15% 1.36% 1.16% Condition

• f migrant

Non recent migrants 100.00% 100.00% 94.01% 0.00% 100.00% Recent migrants 0.00% 0.00% 5.99% 100.00% 0.00% Level of Education No instruction/ Incomplete Elementary School 40.03% 2.49% 25.95% 25.16% 29.91% Complete Elementary School / Incomplete High School 27.68% 4.88% 18.20% 17.56% 22.48% Complete High School / Incomplete College 28.68% 50.05% 32.86% 36.33% 42.59% Complete College 3.61% 42.58% 23.00% 20.96% 5.02% Position of Occupation Formal 100.00% 100.00% 0.00% 99.20% 100.00% Informal 0.00% 0.00% 100.00% 0.80% 0.00% Average monthly income Less than 1 MW 14.13% 0.77% 19.55% 8.41% 7.08% Over 1 to 2 MW 63.87% 30.90% 42.58% 48.61% 67.20% Over 2 to 3 MW 16.30% 11.15% 13.13% 13.61% 17.68% Over 3 to 5 MW 5.54% 19.78% 9.91% 10.76% 5.46% Over 5 to 10 MW 0.15% 23.56% 9.25% 10.44% 2.16% Over 10 MW 0.01% 13.84% 5.57% 8.16% 0.42% Type of Daily Urban Mobility Live and work at the same city 98.70% 89.51% 78.62% 64.63% 8.19% Live at nucleus and work at peripheral city 0.70% 1.96% 1.78% 1.45% 0.19% Live at peripheral city and work at nucleus 0.39% 6.41% 12.27% 22.71% 60.42% Live and work at different peripheral cities 0.21% 2.12% 7.33% 11.20% 31.20% Home-to- work travel time Less than 5 minutes 8.27% 5.39% 7.94% 5.41% 0.22% Over 6 to 30 minutes 40.35% 34.50% 33.52% 32.77% 6.69% Over 31 minutes to 1 hour 34.25% 36.37% 30.78% 30.66% 27.49% Over 1 hour to 2 hours 14.73% 20.37% 22.08% 24.55% 49.20% Over 2 hours 2.39% 3.37% 5.68% 6.62% 16.40% Source: IBGE (2010). Microdata of Sample - 2010 Brazilian Demographic Census