Differences in the spatial landscape of urban mobility: gender and socioeconomic perspectives
Mariana Macedo, Laura Lotero, Alessio Cardillo, Ronaldo Menezes, Hugo Barbosa
DDifferences in the spatial landscape of urbanmobility: gender and socioeconomic perspectives
Mariana Macedo , Laura Lotero , Alessio Cardillo , Ronaldo Menezes , and HugoBarbosa BioComplex Lab, University of Exeter – Exeter, United Kingdom Faculty of Industrial Engineering, Universidad Pontificia Bolivariana – Medell´ın, Colombia Department of Computer Science and Mathematics, University Rovira i Virgili, E-43007 Tarragona, Spain GOTHAM Lab – Institute for Biocomputation and Physics of Complex Systems (BIFI), University of Zaragoza,E-50018 Zaragoza, Spain * [email protected] ABSTRACT
In society, many of our routines and activities are linked to our ability to move; be it commuting to work, shopping for groceries,or meeting friends. Yet, factors that limit the individuals’ ability to fully realise their mobility needs will ultimately affect theopportunities they can have access to (e.g. cultural activities, professional interactions). One important aspect frequentlyoverlooked in human mobility studies is how gender-centred issues can amplify other sources of mobility disadvantages (e.g.socioeconomic inequalities), unevenly affecting the pool of opportunities men and women have access to. In this work, weleverage on a combination of computational, statistical and information-theoretical approaches to investigate the existenceof systematic discrepancies in the mobility diversity (i.e. the diversity of travel destinations) of (1) men and women fromdifferent socioeconomic backgrounds, and (2) work and non-work travels. Our analysis is based on datasets containing multipleinstances of large-scale, official, travel surveys carried out in three major metropolitan areas in South America: Medell´ın andBogot´a in Colombia, and S˜ao Paulo in Brazil. Our results indicate the presence of general discrepancies in the urban mobilitydiversities related to the gender and socioeconomic characteristics of the individuals. Lastly, this paper sheds new light onthe possible origins of gender-level human mobility inequalities, contributing to the general understanding of disaggregatedpatterns in human mobility.
Human travelling behaviours are linked to a myriad of problems in cities such as traffic congestion, disease spreading, andcriminality. Conversely, many of our social and economic activities, such as working, shopping, and socialising hinge on ourability to move. Not surprisingly, human mobility plays a key role in the social and economic development of cities . Oneexample is the economic and social impacts of the mobility restrictions imposed by governments worldwide in 2020 due to theCOVID-19 pandemic . In fact, the economic and social hardships arising from the mobility restrictions unevenly affecteddifferent segments of the populations, exacerbating inequalities of economic, social and gender roots. On the otherhand, the scientific literature in human mobility has vast pieces of evidence indicating the existence of persistent mobilitydifferences across socioeconomic and gender groups . In many instances, urban mobility differences are rooted in theeconomic landscape of a city and the spatial distribution of opportunities (e.g. employment) in urban areas.Thus, understanding the mobility necessities and characteristics of the different segments of the society – especially the lessadvantaged populations – is crucial to reduce social, economic and gender inequalities, objectives contemplated by the UnitedNations in their Sustainable Development Goals .With regards to the socioeconomic facets of urban mobility, in previous works, we have shown that, in Colombia, middle-income populations tend to distribute their visits to most of the areas of a city while upper and lower-income groups are morelikely to concentrate their trips towards a smaller fraction of the zones . Furthermore, it has been shown that in Brazil,populations from different socioeconomic strata tend to use different transportation modes .Moreira et al. suggest that in Brazil, public safety can play an important role in how people move . Even though safety is aproblem faced by all genders, empirical evidences indicate that when possible, women are more likely to opt for longer (or morecostly) journeys in favour of a trip perceived as safer . Nevertheless, in general, women are more likely to make shortertrips than men . Furthermore, women with care duties are more likely to work at locations having shorter commuting travel (last accessed: 9 November 2020) a r X i v : . [ phy s i c s . s o c - ph ] F e b ime, leading them to display different patterns of mobility when compared with men . Hence, the spatial distribution of jobopportunities in a city, combined with the gender division of labour and imbalances in the workloads with care responsibilitiesmay all contribute to gender-centred differences in mobility.Thus, of all the sociodemographic dimensions known to influence human travelling behaviours, in this work, we concentrateon the interaction between gender and the socioeconomic characteristics of travellers and their mobility patterns. We arguethat previously-observed socioeconomic differences in urban mobility could be connected with how different groupsconcentrate/distribute their travels throughout the urban area. Our goal, therefore, is to quantify how concentrated/dispersed thetravelling behaviours of different segments of a population are.Indeed, certain characteristics of the urban areas, combined with the opportunity landscape of the cities, will attract peoplein different ways – and with different magnitudes, – in connection with their sociodemographic characteristics. However, it isnoteworthy that the travelling behaviours of a society are not static in time but, rather, they evolve alongside the population, inresponse to underlying cultural, social and economic changes. Data-driven, longitudinal studies related to sociodemographicsprocesses in human mobility are frequently hindered by data limitations with few exceptions. Using credit card record data,Lenormand et al. showed that women in Spain tend to travel shorter distances, frequently closer to their trajectories’ centre ofmass, while men tend to display longer journeys . We hypothesise that these discrepancies in the mobility patterns of womenand men can be exacerbated when combined with other dimensions such as the socioeconomic status.With this objective in mind, we analyse urban mobility through the lenses of mobility diversity . In our formulation, themobility diversity is measured as the Shannon entropy of the empirical probability distribution of travels made towards theset of zones or sub-areas (e.g., census tracts) of a city. Our analyses are based on multiple waves of household travel surveysfrom three metropolitan areas in South America carried out in different points in time: Medell´ın (2005, 2017) and Bogot´a(2012, 2019) in Colombia, and S˜ao Paulo (1997, 2007, 2017) in Brazil. Each survey is composed of three parts: the travelquestionnaire, focusing on the trips themselves (e.g. destination, modal, and purpose), the household (e.g. number of residentsand family arrangement), and the sociodemographic characteristics of the respondents (e.g. gender, age, and socioeconomicstratum).Our results indicate that, in the areas we analysed, the travel distribution of men and women are marked – consistently –different. Moreover, such differences are not uniform across socioeconomic groups. In fact, the socioeconomic mechanismsoperating on the mobility landscape seem to emphasise and amplify the gender-centred differences in mobility. Our findingsshed new light on the potential mechanisms contributing to gender and socioeconomic disadvantages in urban areas. In abroader perspective, our results suggest that a possible combination of gender biases in employment opportunities with thespecialisation and spatial organisation of the areas of the urban fabric, spurs imbalances in the mobility travel costs sustained bymen and women. In this work, we analyse household travel surveys from three large urban areas in South America, being two in Colombia andone in Brazil. The Colombian datasets correspond to the metropolitan areas of Medell´ın (henceforth stylised as
MDE ) andBogot´a (
BGT ) while the Brazilian dataset covers the metropolitan area of S˜ao Paulo (
SAO ). For each area, we analysed thedata collected in different years: { } for MDE , { } for BGT , and { } for SAO , respectively.Table 1 summarises the main characteristics of each dataset, providing information such as the number of zones covered by thesurveys ( N Z ), their total area ( A ), the number of travellers ( N P ), the number of travels ( N T ), the fractions of travellers ( f X ), andtravels ( f XT ) per gender X ∈ { M , W } . The data of different gender and socioeconomic groups are detailed in Section S1 of theSupplementary Material.The surveys used in this work asked people to describe the set of recent trips they performed. Each travel entry, instead,comes with information on its origin and destination zones, departure and arrival times, the purpose of the travel (also known as demand ), and transportation mode(s) used. Additionally, the surveys also include questions related to the sociodemographiccharacteristics of the respondents, such as their gender, occupation, and socioeconomic status. To account for the representa-tiveness of a respondent’s answers – according to their socioeconomic and demographic characteristics relative to the generalpopulation – each response in the survey is associated with an expansion factor . Such a factor scale up the sample estimating tothe population from which the sample was drawn. We carry out our analysis on the “expanded datasets”, with the sole exceptionof the MDE survey of 2017, for which the expansion factors are unavailable. Notwithstanding, it is worth mentioning that theexpanded datasets are not too different from the original ones (see Section S3.2 of the Supplementary Material). Among theplethora of attributes available to discriminating travellers, we argue that travels related to work purposes are of special interestbecause work activities represent better differences in social strata and gender.To ensure the consistency of our comparative analyses, we harmonised the spatial partitioning of the cities as well as thesocioeconomic categorisation of the respondents. Few zones in each city were split into smaller areas to accommodate changesin their underlying population numbers that occurred between consecutive waves of the survey. Therefore, we decided to use able 1.
Main properties of the raw datasets analysed in our study. For each region, we have the total area covered A , and thenumber of zones into which it is divided, N Z . Then, for each year we have the number of travellers N P , the fraction of men(women) travellers f M ( f W ), the number of travels N T , and the fraction of travels made by men (women) f MT ( f WT ). Region A ( km ) N Z Year N P f M f W N T f MT f WT MDE
BGT
SAO the partitioning corresponding of the first year available in each metropolitan area, and merge together those zones that split inthe following years. We ensured that our aggregation methodology does not alter the overall distributions of travel time, traveldistance, and the fraction of travels. Summing up, throughout our manuscript, the spatial division of the data corresponds tothe area divisions of 2005 (
MDE ), 2012 (
BGT ), and 1997 (
SAO ), respectively. A spatial visualisation of the final divisions canbe seen in Figures 1 and 2, as well as in Section S2 of the Supplementary Material. We also ensured that the socioeconomicclassification of the populations was consistent across years, regions and – to a lesser extent – countries.Ensuring the consistency of socioeconomic status is less straightforward than spatial partitioning. The reason is that notonly the classification might change across time, but also, different countries adopt different criteria/schemes. To interpretthe results for Colombia and Brazil from a common framework, we rearranged the socioeconomic classifications for bothcountries into three socioeconomic strata: lower , middle , and upper . The population distributions obtained from thisrearrangement (Tables S2 and S3 in the Supplementary Material) were similar to what is frequently observed in modernsocieties . Methodological details on the socioeconomic classification of the populations, see Supplementary Material,Section S1.1.After ensuring that the data are aggregated consistently both in terms of spatial partitioning and socioeconomic classificationof the travellers, we can proceed to analyse the mobility patterns across population groups. We decided to study the evolutionin time of the mobility patterns and its similarities/differences between cities using an approach based on information theory.Specifically, we compute a modified version of the mobility diversity indicator proposed by Pappalardo et al. in .Given a set of travels made by a group of travellers X to satisfy/fulfil purpose d , the mobility diversity of such a group, H Xd ,is – up to a multiplicative factor, – the Shannon entropy of its spatial coverage . The latter corresponds to the probability thattravellers from a group X visit a given zone i to satisfy/fulfil purpose d , p Xd ( i ) , yielding: H Xd = − N Z N Z ∑ i = p Xd ( i ) log p Xd ( i ) , (1)where p Xd ( i ) = N Xd ( i ) N Xd . (2)Here, N Xd ( i ) denotes the number of travels made by a group X to fulfil purpose d whose destination is zone i ; whereas N Xd denotes the total number of travels made by a group X to fulfil purpose d . According to Eq. (1), H Xd ∈ [ , ] with the boundaryvalues corresponding to two distinct mobility scenarios. The case H Xd = H Xd =
1, instead, corresponds to the scenario where travels cover uniformly all the availablezones (i.e. Eq. (2) is independent on the zone). The detailed calculations of the boundary values of H are available in SectionS3.1 of the Supplementary Material. Finally, as mentioned previously, the group X can be chosen according to several criteriabased on gender, socioeconomic status, or a combination of them.To account for the effect of variations in sample and population sizes and estimate the variations in mobility diversity inthe populations, we employed a bootstrapping strategy and estimated the H values from random samples of the data. Moreprecisely, given the set of all the travels made by a certain group of travellers, X , fulfilling a given purpose, d , we sample 80% f such travels and then compute the quantity we are interested in (e.g. the value of H Xd using Eq. (1)); we repeat the sampling1000 times. The analyses presented in the next section were performed on the distributions of the mobility diversities obtainedfrom the bootstrapping. From these distributions, we used different statistical methods to verify the differences in the diversitydistributions across groups. Details on the statistical verification methods and results are provided in the SupplementaryMaterial Section S4. In this section, we explore the existence of systematic differences in the mobility patterns of men and women that couldrepresent potential sources of additional disadvantages and inequalities. In line with our hypothesis on the existence of structural,gender-centred mobility disadvantages, we focus our attention on work-related trips. The rationale is that potential genderinequalities permeating the socioeconomic fabric (e.g. employment landscape) would manifest themselves as differences in thecommuting behaviours of men and women, even more so across socioeconomic groups.To explore the role played by gender and socioeconomic factors on urban mobility, we focus our analysis on measuring themobility diversity of the overall travels ( all ) performed by each segment, including, for instance, travels related to shopping , health , and leisure , and their differences to work ( work ) and non-work ( nonwork ) travels. Next, we analyse the mobilitydiversity of the populations across gender or socioeconomic strata. Finally, we investigate the mobility diversity distributionsobtained from the combined effects of both gender and socioeconomic strata. Given that our focus is on the disadvantagesendured by different segments of society, we conducted our last set experiments specifically for the work travels, without furtherpartitioning the data into other travel categories.The visual exploration of the data (Figures 1 and 2, and Section S2 of the Supplementary Material) confirms our hypotheseson the role of gender and socioeconomic status on mobility. We observe, for example, that the majority of the areas in BGT arecovered of a high density of work travels performed by men and by the middle class. As expansions factors are unavailablefor the 2017 survey of
MDE , we are unable to make any claim on the temporal evolution in
MDE . However, in the discussionsection, we will comment about longitudinal changes of H for MDE . Figure 1.
Density map of work travels made in
BGT during the year 2019. Brighter colours represent a higher density oftravels to work. The hue denotes whether for a given zone the majority of travels were made by women (red), men (green), orby both (yellow). The inset portrays a zoom of the city centre. igure 2.
Density map of work travels made in
BGT during the year 2019. Brighter colours represent a higher density oftravels to work. The hue denotes whether for a given zone the majority of travels were made by travellers belonging to the lower (red), middle (green), upper (blue) or all three socioeconomic status. The inset portrays a zoom of the city centre.
To assess the extent to which different purpose of travel shape mobility, we divide the travels into three groups: those related towork ( work ), those related to any purpose except for work ( nonwork ) and, finally, all the travels regardless of their purpose( all ). Then, we compute the mobility diversity, H , of travels belonging to each of the aforementioned groups. Figure 3displays the evolution across time of the distribution of the values of H for each travels’ group and each city. By computing theWelch’s t -test between each pair of distributions, we ensure that they are statistically distinct ( p -value < H are all located above 0.87, meaning that the travels are –more or less – evenly distributed across all the zones available regardless of the purpose, city, or year considered. In general, weobserve that travels of the work group display smaller values of H than those belonging to the other groups. This suggests thatjob opportunities are more spatially concentrated throughout urban areas than other sources of mobility demands together, likeeducation or leisure. Looking at the evolution in time of the diversity of work travels, we observe that both Colombian citiesdisplay an increase of H over time. Such an increase may denote that job opportunities might have appeared in other zones, andthat travels to work became more equally distributed in all the zones. For the city of SAO , instead, we observe an increase in theconcentration of travel destinations from 1997 to 2007 followed by a decrease from 2007 to 2017.
Men and women display different patterns in mobility such as average travel time, preferences on the mode of transportation,and commuting travel distance . Here, we analyse whether mobility diversity is a suitable candidate to graspdifferences in mobility in a gender-centred manner.As an example, we consider the case of
BGT . In Figure 4, we display the Kernel Density Estimation (KDE) of the mobilitydiversity, H , of travels made by men ( M ), women ( W ), and all ( A ) travellers either regardless of the purpose of travel ( all ),and for work travels only ( work ). A quick inspection of Figure 4 reveals that the envelopes of the KDEs tend to get closer(smaller distance between them) and more peaked for the most recent dataset. This means that travellers, regardless of theirgender, tend to choose travel destinations more uniformly over the metropolitan area in 2019 than what they did in 2012. Sucha phenomenon is also corroborated by the values of the peak-to-peak distance between the KDEs of M , W , and A travellersshown in the matrices appearing within each panel.Another feature is that the average values of H obtained for women travellers are always smaller than the same quantitycomputed for the men. Such a comparison between the KDEs confirms that women tend to explore the metropolitan area
005 2017
Year . . . . . . H (a) MDE
Year . . . . . . (b) BGT
Year . . . . . . (c) SAO
Travel type allnonworkwork
Figure 3.
Violin plots of the bootstrapped mobility diversity, H , for all , work and nonwork travels made in each regionand year. To better visualise the overlap (or not) between the distributions of all , work , and nonwork travels, we show thedistributions for all travels duplicated (entire grey violins instead of half-violins).less than men. We ensure that the differences between the KDEs are statistically significant by computing the Welch’s t -testbetween all the possible pairs of distributions ( p -value < BGT , we repeat the sameanalysis also for the data available for
MDE and
SAO (see Section S3.3 of the Supplementary Material).Figure 5 provides an overview of the effects of gender on H for all the urban areas together over all the available years. Evenfor the complete set of areas and time snapshots, the Welch’s t -test confirmed that the distributions are statistically different( p -value < H associated with men’s mobility is higher than that of women regardless of the purpose of travel inagreement with the results observed in Figure 4. The sole exception, however, is the case of SAO in 2007 for which H W > H M .The violin plots show also that, in general, ∆ H MA < ∆ H WA , where ∆ H XY = (cid:12)(cid:12) h H X i − h H Y i (cid:12)(cid:12) and X , Y ∈ { A , M , W } .Focusing on BGT and
SAO , we observe that H increases between 2012 and 2019 regardless of the travel’s purpose or thetraveller’s gender (in Figure 5). The SAO urban area displays the same V-shaped pattern (i.e. the value of H decreases between1997 and 2007, and increases between 2007 and 2017) observed in Figure 3. In particular, we do not notice any qualitativedifferences between the distributions of H all and H work .It is worth mentioning that although in our dataset women are more likely to perform more short travels than men (seeFigure S9 and Tables S4 and S5 of the Supplementary Material) , the lower values of H W do not stem from the preferenceof women to remain within the same zone (see Table S6 of the Supplementary Material). Moreover, our sample does not show ahigh difference in the percentages of men and women living and working in the same zone (see Table S6 of the SupplementaryMaterial). We also investigated whether the number of travel destinations chosen by men is higher than what women choose,but we have not found any statistically significant difference (see Figure S10 of the Supplementary Material). Therefore,individually, women and men have similar likelihoods of performing travel to work in the same number of destinations (zones).The data analysis confirms that the fraction of work travels made by men , P Mwork , is higher than its women counterpart. Onthe other hand, we have found that non-work related travels are proportionally higher for women than men (see Table S7 of theSupplementary Material). Moreover, except for
MDE in 2017, the travel’s destination for women and men follows differentdistributions regardless of the purpose of travel (tested by Student t -test and Kolmogorov–Smirnov test with p -value < . H displayed in Figure 5, and the balance between genders in the compositionof travellers’ groups, push us to ask whether such differences are concealed by other factors related, for instance, with thesocioeconomic status of travellers. For this reason, we study the effects of socioeconomic status in mobility. Finally, we explore the effect of socioeconomic status and gender in mobility diversity. However, before studying the effects ofthese two aspects combined, we must gauge the role of socioeconomic status alone. For this reason, we grouped travellersaccording to the three socioeconomic classes defined in Section 2 (i.e. lower
Low , middle Mid , and upper
U p ). Wecomputed the values of H of travels made by travellers belonging to each socioeconomic class, as well as for travels made byall travellers combined ( A ), and for travels made for either all or work purposes.In Figure 6, we display the KDE ( H ) for the BGT area for the years 2012 and 2019, respectively. In agreement with thetrend observed thus far, we observe an increase over time of the mobility diversity independently on the purpose of travel .
88 0 .
90 0 .
92 0 . a ll t r a v e l s K D E ( H ) (a) A M WAMW −− − . − . . .
88 0 .
90 0 .
92 0 . (b) A M WAMW −− − . − . . .
88 0 .
90 0 .
92 0 . H w o r k t r a v e l s K D E ( H ) (c) A M WAMW −− . − . . .
88 0 .
90 0 .
92 0 . H (d) A M WAMW −− . − . . All (A)
Men (M)
Women (W)
Figure 4.
Kernel Density Estimation plots of the mobility diversity, H , for all travels (panels a and b ) and work travels(panels c and d ) in the urban area of BGT . For each travel purpose, we plot KDE ( H ) for travels made by men ( M ), women ( W ),and all travellers ( A ). The matrix appearing within each graphic summarises the distances between the medians of thedistributions (peak-to-peak distances multiplied by 10 − ). H a ll t r a v e l s (a) MDE 0.860.880.900.920.94 (b)
BGT 0.860.880.900.920.94 (c)
SAO
Year H w o r k t r a v e l s (d) Year (e)
Year (f)
All (A)Men (M)Women (W)
Figure 5.
Violin plots of the bootstrapped mobility diversity, H , for all travels (top row, panels a-c), and work travels(bottom row, panels d-f). Each column refers to a different region: MDE (panels a and d),
BGT (panels b and e), and
SAO (panels c and f). For each region, we display the distribution of the values of H in each year. We show the distributions for all travels duplicated (entire violins), and the distributions for men and women travels in half-violins. . . . a ll t r a v e l s K D E ( H ) (a) A Low Mid UpALowMidUp −−− . −− . − . − . . . . . . (b) A Low Mid UpALowMidUp −−− . −− . − . − . . . . . . H w o r k t r a v e l s K D E ( H ) (c) A Low Mid UpALowMidUp −−− . −− . . − . . . . . . H (d) A Low Mid UpALowMidUp −−− . −− . − . − . . . All (A)
Lower (Low)
Middle (Mid)
Upper (Up)
Figure 6.
Kernel Density Estimation (KDE) plots of the mobility diversity, H , for travels made by travellers belonging todifferent socioeconomic status in the BGT area of 2012 (panels a and c) and 2019 (panels b and d). The top row (panels a and b)displays the values obtained considering all travels, whereas the bottom row (panels c and d) displays the values obtainedconsidering only travels associated with the work purpose. We show the KDE ( H ) for travels made by all travellers ( A ), as wellas for those belonging to the lower ( Low ), middle ( Mid ), or upper ( U p ) socioeconomic class. The matrix appearingwithin each plot encapsulates the distance between the median of the distributions (peak-to-peak distances multiplied by 10 − ).considered. Travellers belonging to the upper class attain the lowest value of H , suggesting that they cover less uniformly thespace. In other words, upper-income individuals might be more “selective” in their mobility than those belonging to the others.On the other hand, in general, middle class travellers display the highest values of H . The higher values of H Mid over theother classes suggest that individuals in the middle class cover the space more uniformly than other classes. However, thereasons for such heterogeneity in the upper and lower classes are not the same. People belonging to upper-income classmove to fewer zones because they probably do not need to seek for opportunities in other neighbourhoods. People belonging tolower-income class, instead, move to fewer zones because they cannot afford to reach all of them. For instance, they mightnot be able to reach urban areas where the public transportation system is insufficient or inadequate. Finally, we observe thatthe peak-to-peak distances between the KDE become smaller over the years, indicating a possible decrease of socioeconomicinequalities in
BGT .Regarding the other urban areas, we observe that the KDE plots (see Figures S11 and S12 of the Supplementary Material)confirm that travellers belonging to the upper class attain the lowest values of H , whereas those belonging to the middle class cover more uniformly the available space. Such socioeconomic magnitudes of H do not depend on the travel’s purpose,albeit each urban area displays its own peculiarities.After analysing the role of socioeconomic status alone, we are now ready to look at the combined effect of gender andsocioeconomic status. To this aim, we compute the mobility diversity of travels made by travellers having a certain social status(e.g. middle ) and gender (e.g. W ). In Figure 7, we display the violin plots of H computed for travels made for all purposesby all combinations of gender and socioeconomic status.First, we noticed that socioeconomic status shapes the mobility of people considerably, whereas gender exerts a smallereffect. Yet, we can observe a gender distinction, with men tending to display higher values of H than women within the samesocioeconomic class. On average, the gender differences within each class tend to diminish over time, suggesting that thegender gap might be getting smaller (see Table S8 of the Supplementary Material). We noticed that the starker differencesbetween genders occur for travellers belonging to the upper class (see Table S8 of the Supplementary Material). Similarconclusions can be drawn from the mobility diversity computed for travels made for work purposes by all combinations ofgender and socioeconomic status (see Figure S13 of the Supplementary Material). We highlight that we can not reject the null ypothesis that the distributions of the work travels are statistically similar (tested by Welch’s t -test with p -value < .
01) inthree cases: (i) comparing women and men from the upper class of
BGT in 2019; (ii) comparing women and men from the middle class of
SAO in 1997; and (iii) comparing men from the lower and middle class of
MDE in 2005. H (a) MDE (b)
BGT (c)
SAO
All travels
AllMen Lower classWomen Lower classMen Middle classWomen Middle classMen Upper classWomen Upper class
Figure 7.
Violin plots of the mobility diversity, H , of travels made for all purposes by travellers grouped according to theirsocioeconomic status and gender. Each column refers to a different region, and for each region, we consider all the availableyears. For each socioeconomic status ( upper , middle , and lower ) a darker hue denotes men travellers, whereas lighter huedenotes women ones.To assess the contribution of the gender and socioeconomic attributes (alone and combined) in the mobility diversity H ,we apply three statistical tests. We, first, apply the ANOVA one-way test to investigate further if the averages of the mobilitydiversity distributions computed separately by either the gender or socioeconomic status groups are statistically different. Then,we apply the ANOVA two-way test to investigate if the averages of the mobility diversity distributions computed by genderand socioeconomic status together are statistically different. Then, we apply the Tukey’s HSD post hoc test to identify withinattributes what are the groups with statistically different average values of H. All the detailed explanation and specific values of F and p -values from ANOVA and Tukey’s HSD post hoc tests are detailed in Section S4 of the Supplementary Material.Based on the three statistical tests, we can reject the hypothesis that the mean values of the mobility diversity, H , from thetravels performed by gender (men, women and all travellers) and socioeconomic groups ( lower , middle , upper and alltravellers) are similar. When considering only the gender or the socioeconomic status, there is no exception in the statisticaltests. astly, we compare the mobility diversity distributions of gender and socioeconomic status taken together. We rejectthe null hypothesis that the mean values of the mobility diversity, H , distributions are similar from the majority of pairwisecomparisons, except for the work travels performed by (i) men from lower and middle classes of MDE in 2005; (ii) men and women from middle class of
SAO in 1997.Summing up, in general, the patterns in mobility diversity for different groups of gender and socioeconomic classes takenseparately or together are statistically different. Without exception, we can claim that the socioeconomic group consistentlyaccounts for the highest gap in mobility diversity.
In search to understand general patterns – universalities – in human mobility, a consistent body of literature assumes thattravellers are indistinguishable from one another . However, travellers are different, and they can be differentiated accordingto several features. To this aim, we analyse the travel records collected by surveys conducted across several years in twoColombian (Medell´ın and Bogot´a) and one Brazilian (S˜ao Paulo) metropolitan areas. We demonstrate that features like genderand socioeconomic status exert a strong influence on the individuals’ mobility patterns at the urban level/scale.Using information theory, we measured the spatial diversity of the travels performed by different groups of people belongingto different gender and socioeconomic groups through a modified version of Shannon’s entropy. Such a quantity, namedmobility diversity, depends on where the travels take place, i.e. the probability that a zone is the travel destination. Thus,mobility diversity can be thought of as a proxy of the “predictability” of a group’s mobility .The travel records collected by the surveys come with meta-data such as age, gender, socioeconomic status, and familyrelationships. Such attributes may be used to group travels (and travellers) according to several criteria. We decided to focus onthree class of attributes: the purpose of the travel, gender of the traveller, and his/her socioeconomic status. To decipher howeach attribute shapes the mobility, we analysed the role of each attribute alone first, and then the role of the attributes takenaltogether.The literature shows that the purpose of travels (e.g. going home and to work) shapes mobility differently in severalspatio-temporal characteristics such as the probability of returning to the last visited location, the fraction of travels over time,the most frequent visited locations, and the amount of money spent related to the distance travelled . To unveil the roleplayed by the purpose of the travel, we have divided travels into three groups/classes: one made of travels due to work activities( work ), one made of travels due to any purpose except work ( nonwork ), and another made by all travels together regardlessof their purpose ( all ). We have found that work-related travels – in general – are less homogeneously distributed than the othertypes of travels. This could be due to the fact that areas that offer a higher amount of job opportunities concentrate a highernumber of travels related to work. However, we have observed that each urban area evolves differently over time, suggestingthat mobility is strongly intertwined with the economic context where it takes place. Thus, the patterns observed in the presentwork cannot be considered universal.The analysis of the role played by gender has revealed the existence of a distinction between the mobility of men andwomen, with the former being more entropic/diverse than the latter. In our analysis, such a phenomenon is independent ofthe region, time, and purpose of the travel. Such a difference between genders has also been highlighted by other studies onmobility which have found, among other things, that women tend to make shorter travels than men, and avoidto travel to certain destinations particularly during late hours. Moreover, we have observed that the gender differences inmobility diversity get smaller over time. Although each area/country has undergone different financial and social changes,such a reduction might be due to the effects of policies aimed either at reducing the gender gap directly or mitigating factorshindered on women’s mobility (e.g. insecurity) .When it comes to the role of the socioeconomic status, travellers can be analysed as a group. Our analysis highlights thatthe diversity gap between socioeconomic classes (SESs) is starker than the gender case. Both the differences between thevalues of H attained by each SES, and the overall range of values of H , point to wealth playing a crucial role in shaping theurban mobility. Moreover, we observe a distinction among SESs, with upper and lower being the classes exploring thespace in the least diverse way, and middle travellers displaying the most diverse mobility patterns . However, the reasonsbehind the lower values of H attained by upper and lower are not the same. The upper class, in fact, appears to be moreselective in their destinations (and also move less) possibly because they can afford a broad range of options (i.e. buying a caror living in expensive areas closer to where they work). The lower class, instead, limit their exploration of the available spacebecause they lack affordable ways to move across the metropolitan areas .Finally, we examined the effect of combining gender and socioeconomic status in one analysis. Although SES plays a majorrole in discriminating travellers, we noticed that each SES displays a further separation based on gender, with men attaininghigher values of H than women. Such separation is independent on the purpose of travel, region, and year. Hence, we canconclude that the gender gap in mobility is a widespread phenomenon affecting women tout-court . In fact, regardless the region, he upper class displays the highest gender difference, and this may be true because there is a higher gender inequality in highlyqualified jobs that women are less likely to pursue .Past studies concluded that features like gender and age of the travellers do not play a remarkable role in our ability to predicttheir mobility behaviour . More recent studies, instead, have highlighted that gender plays a role in the discriminationof travellers . Although some studies have addressed the role of the gender and the socioeconomic status of travellersseparately ; to the best of our knowledge, their combined effect has been studied seldom. We presented here asystematic study of how gender and socioeconomic status are intertwined and shape urban mobility. In particular, we observedthat there is a gap between gender regardless of the socioeconomic status of the traveller. In fact, women report disadvantagesin several aspects of their life such as income, free time, and career’s progression , and this work shows that mobility (andaccess to transport) is another aspect in which women suffer .Nevertheless, our analysis has some limitations. The first is that results are intimately tied to the type of data used. Despitebeing very detailed and rich in meta-data, surveys are very expensive to carry out (both economically and time-wise). Moreover,the information collected could be exposed to subjective biases on both the interviewer and the interviewed sides. Otherlimitations include the composition of the sample, and the tendency to capture only routine behaviours. Expansion factorssurely mitigate the aforementioned issues, but they are not always available. Finally, the spatio-temporal resolution/accuracy ofsurveys is not comparable with other types of data such as mobile phones or credit card records.Future extensions of our work could involve the study of other travel purposes (e.g. related to study or leisure), or theeffects of the traveller’s age/profession on its mobility pattern. Finally, expanding the pool of countries/cities, including casesfrom countries located outside Latin America, could help to unveil more generalised trends on the gender gap in urban mobility. References Xu, Y., Belyi, A., Bojic, I. & Ratti, C. Human mobility and socioeconomic status: Analysis of singapore and boston.
Comput. Environ. Urban Syst. , 51–67 (2018). Bettencourt, L. M. The origins of scaling in cities. science , 1438–1441 (2013). Barbosa, H. et al.
Uncovering the socioeconomic facets of human mobility. arXiv preprint arXiv:2012.00838 (2020). Warren, M. S. & Skillman, S. W. Mobility Changes in Response to COVID-19. arXiv (2020). 2003.14228. Bonaccorsi, G. et al.
Economic and social consequences of human mobility restrictions under COVID-19.
Proc. Natl.Acad. Sci. , 15530–15535, DOI: 10.1073/pnas.2007658117 (2020). Ho, K. W., Kheng, M. & Tan, T. Challenges to social mobility in singapore.
SSRN
DOI: 10.2139/ssrn.3643266 (2020). Kraemer, M. U. et al.
The effect of human mobility and control measures on the COVID-19 epidemic in China.
Science , 493–497, DOI: 10.1126/science.abb4218 (2020). Coven, J. & Gupta, A. Disparities in mobility responses to COVID-19. Tech. Rep., NYU Stern Working Paper (2020). Blundell, R., Costa Dias, M., Joyce, R. & Xu, X. COVID-19 and Inequalities.
Fiscal Stud. , 291–319, DOI: 10.1111/1475-5890.12232 (2020). Collins, C., Landivar, L. C., Ruppanner, L. & Scarborough, W. J. COVID-19 and the gender gap in work hours.
Gender,Work. & Organ.
Myers, K. R. et al.
Unequal effects of the COVID-19 pandemic on scientists.
Nat. Hum. Behav. , 880–883, DOI:10.1038/s41562-020-0921-y. Salon, D. & Gulyani, S. Mobility, poverty, and gender: travel ‘choices’ of slum residents in nairobi, kenya.
Transp. Rev. , 641–657 (2010). Adeel, M., Yeh, A. G. & Zhang, F. Gender inequality in mobility and mode choice in pakistan.
Transportation ,1519–1534 (2017). Lotero, L., Cardillo, A., Hurtado, R. & G´omez-Garde˜nes, J. Several Multiplexes in the Same City: The Role of Socioe-conomic Differences in Urban Mobility. In Garas, A. (ed.)
Interconnected Networks. Understanding Complex Systems ,Understanding Complex Systems, 149–164, DOI: 10.1007/978-3-319-23947-7 9 (Springer International Publishing, Cham,2016).
Lotero, L., Hurtado, R. G., Flor´ıa, L. M. & G´omez-Garde˜nes, J. Rich do not rise early: spatio-temporal patterns in themobility networks of different socio-economic classes.
Royal Soc. open science , 150654, DOI: 10.1098/rsos.150654(2016). Vasconcellos, E. A. Urban transport policies in Brazil: The creation of a discriminatory mobility system.
J. transportgeography , 85–91, DOI: 10.1016/j.jtrangeo.2017.08.014 (2018). Moreira, G. C. & Ceccato, V. A. Gendered mobility and violence in the S˜ao Paulo metro, Brazil.
Urban Stud.
Mej´ıa-Dorantes, L. & Soto Villagr´an, P. A review on the influence of barriers on gender equality to access the city: Asynthesis approach of Mexico City and its Metropolitan Areaa.
Cities , 102439, DOI: 10.1016/j.cities.2019.102439(2020). Ng, W.-S. & Acker, A. Understanding urban travel behaviour by gender for efficient and equitable transport policies(International Transport Forum Discussion Paper, 2018).
Singh, Y. J. Is smart mobility also gender-smart?
J. Gend. Stud. , 832–846, DOI: 10.1080/09589236.2019.1650728(2020). Macedo, M., Lotero, L., Cardillo, A., Barbosa, H. & Menezes, R. Gender patterns of human mobility in Colombia:Reexamining Ravenstein’s laws of migration. In
Complex Networks XI: Proceedings of the 11th Conference on ComplexNetworks CompleNet 2020 , 269–281, DOI: 10.1007/978-3-030-40943-2 23 (Springer Nature, 2020).
Stoet, G. & Geary, D. C. A simplified approach to measuring national gender inequality.
PLoS ONE , 1–18, DOI:10.1371/journal.pone.0205349 (2019). Kaufman, G. & Taniguchi, H. Gender equality and work–family conflict from a cross-national perspective.
Int. J. Comp.Sociol. , 385–408, DOI: 10.1177/0020715219893750 (2020). Petrongolo, B. & Ronchi, M. Gender gaps and the structure of local labor markets.
Labour Econ. , 101819, DOI:10.1016/j.labeco.2020.101819 (2020). Lenormand, M. et al.
Influence of sociodemographic characteristics on human mobility.
Sci. Reports , 10075, DOI:10.1038/srep10075 (2015). Pappalardo, L., Pedreschi, D., Smoreda, Z. & Giannotti, F. Using big data to study the link between human mobilityand socio-economic development.
Proc. - 2015 IEEE Int. Conf. on Big Data, IEEE Big Data 2015
Pappalardo, L. et al.
An analytical framework to nowcast well-being using mobile phone data.
Int. J. Data Sci. Anal. ,75–92, DOI: 10.1007/s41060-016-0013-2 (2016). 1606.06279. Leo, Y., Fleury, E., Alvarez-Hamelin, J. I., Sarraute, C. & Karsai, M. Socioeconomic correlations and stratification insocial-communication networks.
J. The Royal Soc. Interface , 20160598, DOI: 10.1098/rsif.2016.0598 (2016). Gao, J., Zhang, Y. C. & Zhou, T. Computational socioeconomics.
Phys. Reports , 1–104, DOI: 10.1016/j.physrep.2019.05.002 (2019).
Gustafson, P. Work-related travel, gender and family obligations.
Work. employment society , 513–530, DOI: 10.1177/0950017006066999 (2006). Rodr´ıguez, L., Palanca, J., del Val, E. & Rebollo, M. Analyzing urban mobility paths based on users’ activity in socialnetworks.
Futur. Gener. Comput. Syst. , 333–346, DOI: 10.1016/j.future.2019.07.072 (2020).
Psylla, I., Sapiezynski, P., Mones, E. & Lehmann, S. The role of gender in social network organization.
PLoS ONE ,1–21, DOI: 10.1371/journal.pone.0189873 (2017). Gauvin, L. et al.
Gender gaps in urban mobility.
Humanit. Soc. Sci. Commun. , 1–13, DOI: 10.1057/s41599-020-0500-x(2020). Gonzalez, M. C., Hidalgo, C. A. & Barabasi, A.-L. Understanding individual human mobility patterns.
Nature , 779,DOI: 10.1038/nature06958 (2008).
Simini, F., Gonz´alez, M. C., Maritan, A. & Barab´asi, A.-L. A universal model for mobility and migration patterns.
Nature , 96, DOI: 10.1038/nature10856 (2012).
Asgari, F., Gauthier, V. & Becker, M. A survey on Human Mobility and its applications. arXiv (2013). 1307.0814.
Barbosa, H., de Lima-Neto, F. B., Evsukoff, A. & Menezes, R. The effect of recency to human mobility.
EPJ Data Sci. ,21, DOI: epjds/s13688-015-0059-8 (2015). Barbosa, H. et al.
Human mobility: Models and applications.
Phys. Reports , 1–74, DOI: 10.1016/j.physrep.2018.01.001(2018). Alessandretti, L., Aslak, U. & Lehmann, S. The scales of human mobility.
Nature , 402–407 (2020).
Song, C., Qu, Z., Blumm, N. & Barab´asi, A.-L. Limits of predictability in human mobility.
Science , 1018–1021, DOI:10.1126/science.1177170 (2010).
Ballis, H. & Dimitriou, L. Revealing personal activities schedules from synthesizing multi-period origin-destinationmatrices.
Transp. research part B: methodological , 224–258, DOI: 10.1016/j.trb.2020.06.007 (2020).
Lenormand, M., Arias, J. M., San Miguel, M. & Ramasco, J. J. On the importance of trip destination for modellingindividual human mobility patterns.
J. The Royal Soc. Interface , 20200673, DOI: 10.1098/rsif.2020.0673 (2020). Farr´e, L., Jofre-Monseny, J. & Torrecillas, J. Commuting Time and the Gender Gap in Labor Market Participation.
IZADiscuss. Pap.
Montoya-Robledo, V. et al.
Gender stereotypes affecting active mobility of care in Bogot´a.
Transp. Res. Part D: Transp.Environ. , 102470, DOI: 10.1016/j.trd.2020.102470 (2020). Chant, S. Cities through a ”gender lens”: A golden ”urban age” for women in the global South?
Environ. Urbanization ,9–29, DOI: 10.1177/0956247813477809 (2013). Carlsson-Kanyama, A. & Linden, A.-L. Travel patterns and environmental effects now and in the future:: implications ofdifferences in energy consumption among socio-economic groups.
Ecol. Econ. , 405–417, DOI: 10.1016/S0921-8009(99)00006-3 (1999). Carra, G., Mulalic, I., Fosgerau, M. & Barthelemy, M. Modelling the relation between income and commuting distance.
J.Royal Soc. Interface , DOI: 10.1098/rsif.2016.0306 (2016). Alonso, W. Location and land use. toward a general theory of land rent.
Harv. Univ. Press. (1964).
Frias-Martinez, V., Virseda-Jerez, J. & Frias-Martinez, E. On the relation between socio-economic status and physicalmobility.
Inf. Technol. for Dev. , 91–106, DOI: 10.1080/02681102.2011.630312 (2012). De Vasconcellos, E. A. Urban change, mobility and transport in S˜ao Paulo: three decades, three cities.
Transp. Policy ,91–104, DOI: 10.1016/j.tranpol.2004.12.001 (2005). Stier, H. & Yaish, M. Occupational segregation and gender inequality in job quality: a multi-level approach.
Work.employment society , 225–246, DOI: 10.1177/0950017013510758 (2014). Clarke, H. M. Gender stereotypes and gender-typed work. In
Handbook of Labor, Human Resources and PopulationEconomics , 1–23, DOI: 10.1007/978-3-319-57365-6 21-1 (Springer, 2020).
Alessandretti, L., Sapiezynski, P., Sekara, V., Lehmann, S. & Baronchelli, A. Evidence for a conserved quantity in humanmobility.
Nat. Hum. Behav. , 485–491, DOI: 10.1038/s41562-018-0364-x (2018). Martin, F. & Dragojlovic, A. Gender, Mobility Regimes, and Social Transformation in Asia.
J. Intercult. Stud. , 275–286,DOI: 10.1080/07256868.2019.1599166 (2019). Levin, L. & Thoresson, K. Gender equality and “smart” mobility.
Gendering Smart Mobilities
Datos abiertos del area metropolitana de valle de aburr´a. Available at: https://datosabiertos.metropol.gov.co/search/field topic/movilidad-y-transporte-2/type/dataset?sort by=changed.
List of abbreviations
MDE : The urban area of Medell´ın.
BGT : The urban area of Bogot´a.
SAO : The urban area of S˜ao Paulo. all : The whole travels available. work : Travels related with work activities. nonwork : Travels not related with work activities. omen : Travels performed by women. men : Travels performed by men.
SES:
Socioeconomic status. lower : Travels performed by travellers belonging to the lower socioeconomic class. middle : Travels performed by travellers belonging to the middle socioeconomic class. upper : Travels performed by travellers belonging to the upper socioeconomic class.
Competing interests
The authors declare that they have no competing interests.
Author’s contributions
MM and HB designed the study; LL and MM contributed with the data; MM and HB performed the analysis; MM and HBanalysed the results; MM and AC wrote the paper. MM and AC prepared the graphics. All authors read, reviewed, and approvedthe final manuscript.
Availability of data and materials
All the data analysed in this paper are publicly available online: Medell´ın and Bogot´a and S˜ao Paulo . Resources will bealso available on github.com/marianagmmacedo/differences_urban_mobility . Acknowledgements
We thank Area Metropolitana del Valle de Aburr´a, in Medell´ın, and Secretarıa Distrital de Movilidad, in Bogot´a, for theOrigin-Destination Surveys Datasets. This work was made possible by a Visiting Fellowship of LL in the Leicester Institute forAdvanced Studies at the University of Leicester.
Funding
RM, HB and MM acknowledge support in part by the U. S. Army Research Office (ARO) under grant number W911NF-18-1-0421. LL acknowledges partial support from the Leicester Institute for Advanced Studies under the Rutherford FellowshipScheme funded by the Department for Business, Energy and Industrial Strategy, and administered by Universities UKInternational. AC acknowledges the support of the Spanish Ministerio de Ciencia e Innovaci´on (MICINN) through GrantIJCI-2017-34300.auto-ignore upplementary Materials for the manuscript entitled:Differences in the spatial landscape of urban mobility: gender andsocioeconomic perspectives
Mariana Macedo, Laura Lotero, Alessio Cardillo, Ronaldo Menezes, and Hugo Barbosa
S1 The mobility surveys
As described in the main manuscript, in this work, we analyse the data collected from travel surveys carried out inthree large South American urban areas: two in Colombia and one in Brazil. The Colombian datasets correspondto the metropolitan area surrounding the city of Medell´ın (henceforth indicated as
MDE ), and the metropolitan areaof Bogot´a (
BGT ). The Brazilian dataset corresponds to the mobility taking place in the metropolitan area of S˜aoPaulo (
SAO ). For each area, we analysed the data collected in different years: { } for MDE , { } for BGT , and { } for SAO , respectively.
S1.1 Harmonising the socioeconomic classification across years and cities
Despite the fact that Brazil and Colombia are both developing countries from South America, the socioeconomiccharacteristics of the three cities and their populations are different. Furthermore, at the time scale of the travelsurveys, there are significant economic changes even at a city level. Thus, one important step in our analyses isharmonising the socioeconomic classification across years, cities and countries.For the Colombian datasets, the socioeconomic classification of the respondents is kept consistent across yearsand cities. Households are split into six strata, and this classification has been widely used as a proxy of thesocioeconomic status of individuals, with stratum corresponding to people with the lowest income, and stratum corresponding to people with the highest income, instead. The mapping between the aforementioned strata andour partition is: lower (strata and ), middle (strata and ), and upper (strata and ), respectively.In our data for the city of S˜ao Paulo, however, the socioeconomic classification of the population is basedon the methodological standards adopted by the Brazilian census authority and their socio-demographic researchinstitute at the time of the survey. Not surprisingly, the classification methodology changes over time to bettercapture a current picture of the socioeconomic characteristics of the population. More precisely, for the 1997 data,respondents were classified into five socio-economic classes labelled as A (upper), B , (mid-upper), C (middle), D (mid-lower) and E (lower). It is noteworthy that this division takes into account not only their overall incomesbut other characteristics such as standard of living, purchase power, housing conditions, and access to amenitiesand transport infrastructure. More recently, Brazilian institutes such as IBGE (Brazilian geography and statisticsinstitute) and ABEP (Brazilian association for population studies) adopted sub-divisions of these major groups toprovide a more precise picture of the population’s realities in terms of their socioeconomic statuses. The 2007and 2017 S˜ao Paulo’s travel survey data also utilised these subdivisions. The division we adopted in terms of ourpartition is presented in Table S1.Supplementary Table S1: Mapping of the Brazilian classification scheme into the lower , middle , and upper socioeconomic classes (SES) for the three years of the survey. YearSES 1997 2007 2017 lower D , E C2 , D , E C2 , D , E middle B , C B1 , B2 , C1 B1 , B2 , C1 upper A A1 , A2 A a r X i v : . [ phy s i c s . s o c - ph ] F e b In this section, we provide a general overview of our datasets and their compositions in terms of their numbersof underlying populations and their travels. We also provide their partition across the socioeconomic and gen-der dimensions. Table S2 displays the composition of the complete datasets whereas Table S3 reports the samequantities for the subsets containing the work travels only. In both tables, quantities denoted with the symbol N represent counts, while quantities denoted with the symbol f represent fractions corresponding to distinct groups.Furthermore, in our notation, the superscript text refers to the group, and a subscript T is used whenever we referto the travels. Notice that these fractions are computed using the expanded data, meaning that they are not relativeto our sample sizes but rather to how many people/travels they represent.Supplementary Table S2: Summary of the composition of all the expanded data sets for travels made for all purposes. For a given location and year, we report: the number of travellers N P , the number of travels N T , thefraction of men (women) travellers f M ( f W ), and the fraction of travels made by men (women) f MT ( f WT ). Wereport also the fraction of travellers belonging to the lower ( f lower ), middle ( f middle ), and upper ( f upper )socioeconomic classes, and the same quantities discriminated by gender (e.g. f lower W ). Finally, we report thefraction of travels made by travellers with a given socioeconomic class and gender (e.g. f lower WT ). The data setsare obtained applying the expansion factors to the raw data from the surveys. Location Medell´ın (
MDE ) Bogot´a (
BGT ) S˜ao Paulo (
SAO )Year 2005 2017 2012 2019 1997 2007 2017 N P N T f M f W f MT f WT f lower f middle f upper f lower T f middle T f upper T f lower M f middle M f upper M f lower W f middle W f upper W f lower MT f middle MT f upper MT f lower WT f middle WT f upper WT work purpose. See the caption of Table S2 for the description of each row. Location Medell´ın (
MDE ) Bogot´a (
BGT ) S˜ao Paulo (
SAO )Year 2005 2017 2012 2019 1997 2007 2017 N P N T f M f W f MT f WT f lower f middle f upper f lower T f middle T f upper T f lower M f middle M f upper M f lower W f middle W f upper W f lower MT f middle MT f upper MT f lower WT f middle WT f upper WT S2 Spatial distribution of travels and their population compositions
Here, we provide some additional visual insights to the underlying composition of the travellers by means ofdensity maps. In a density map, thousands of points of different colours are scattered within each area. Thenumber of points is proportional to a measure of interest, whereas their colours encode the groups they belong to.Such an encoding means that denser areas will appear brighter in the map, while the group composition will bereflected on the colour of the area. In our case, the number of points in an area is proportional to the number of work travels having each area (or zone) as their destination, whereas the colours correspond to either the genderor the socioeconomic groups to which the travellers belong. It is noteworthy that density maps are not intendedto provide an accurate, quantitative representation of the population compositions but, rather, to give an overallperspective on the spatial distribution of the trips in terms of their density, mixing, and segregation. For brevity,here we show the visualisations only for the most recent data for each city.
S2.1 Gender composition
We looked at the gender composition of the work-related travels for the cities of
MDE and
SAO (Figures S1 andS2) respectively. First, in both cities, it is evident the presence of a larger concentration of travels in their central3reas. Furthermore, we can see also that in the centre of the cities, the work travels are more gender-balanced,hence the predominance of brighter white zones. However, some less dense areas exhibit small fluctuations intheir gender balances, with a slight prevalence of areas coloured in green. Therefore, it is evident that the originsof the significant differences in the number of work travels made by men and women reported in Table S3 comefrom the less dense areas of the cities.Supplementary Figure S1: Density map of work travels made in
MDE during the year 2017. Brighter coloursrepresent a higher density of travels to work. The hue denotes whether for a given zone the majority of travelswere made by women (red), men (green), or by both (yellow). The inset portrays a zoom of the city centre.Supplementary Figure S2: Density map of work travels made in
SAO during the year 2017. Brighter coloursrepresent a higher density of travels to work. The hue denotes whether for a given zone the majority of travelswere made by women (red), men (green), or by both (yellow). The inset portrays a zoom of the city centre.4
Another way to look at the travel distribution is through their socioeconomic compositions. One striking featureobserved in both Medell´ın (Figure S3) and S˜ao Paulo (Figure S4) is that the more visited areas of the cities arealso homogeneous with regards to the socioeconomic characteristics of their visiting populations. This is causedby the fact that the central districts of these cities tend to concentrate a large portion of their economic activitiesand businesses, therefore attracting workers from a broader range of segments, sectors, and backgrounds.Despite these marked socioeconomic homogeneities at the centre, we can also observe in Figures S3 and S4that there are indeed areas incline to attract more predominantly workers from specific socioeconomic groups.Both in Medell´ın and S˜ao Paulo, it is possible to observe areas coloured in red, indicating a stronger concentrationof work travels by lower-income people. Additionally, outside the dense core of the cities, we can observe thatboth cities tend to have areas that seem to be less attractive to specific income groups. For instance, in Medell´ın,most of the areas are coloured in yellow shades, indicating that those zones attract more workers of lower andmiddle income and less of upper income. A similar pattern can also be observed, albeit in a lesser extent, in S˜aoPaulo. In fact, S˜ao Paulo tends to have more zones coloured with blue and red hues than Medell´ın, suggesting thatS˜ao Paulo is a city in which the economic landscape tend to be more segregated .Supplementary Figure S3: Density map of work travels made in
MDE during the year 2017. Brighter coloursrepresent a higher density of travels to work. The hue denotes whether for a given zone the majority of travelswere made by travellers belonging to the lower (red), middle (green), upper (blue) or all three socioeconomicstatus. The inset portrays a zoom of the city centre. 5upplementary Figure S4: Density map of work travels made in
SAO during the year 2017. Brighter coloursrepresent a higher density of travels to work. The hue denotes whether for a given zone the majority of travelswere made by travellers belonging to the lower (red), middle (green), upper (blue) or all three socioeconomicstatus. The inset portrays a zoom of the city centre.
S3 Mobility diversity
S3.1 Boundary values of the mobility diversity
Following Eq. (1), one could demonstrate that the mobility diversity of a group of travellers X travelling to fulfilpurpose d is bounded (i.e. H Xd ∈ [0 , ). Such boundary values have a clear, physical, meaning which is relatedto the characteristics of the probability that travels have as their destination a given zone i , p Xd ( i ) , presented inEq. (2). In the following, we compute the boundary values. Noteworthy, these boundaries do not depend on eitherthe group of travellers, X or the purpose of travel, d , under consideration.The least diverse mobility pattern corresponds to the case where travellers travel exclusively to one zone (say, i = ˜ i ). Under such an assumption, Eq. (2) becomes: p Xd ( i ) = ( for i = ˜ i otherwise . (S1)By replacing p in Eq. (1), H Xd reads: H Xd = − N Z " (1 log
1) + N Z X i =1 i =˜ i = − N Z (0 + 0) = 0 . (S2)If, instead, we assume that the travellers cover all the available zones uniformly, then each destination is reachedby the same number of travels, corresponding to the most diverse mobility pattern. Under such circumstances,Eq. (2) becomes: p Xd ( i ) = N Xd ( i ) N Xd = N Xd /N Z N Xd = N Xd N Z N Xd = 1 N Z ∀ i , (S3)where N Xd ( i ) is the total number of trips made by a group X with a purpose d to a destination area i and N Xd =6 N Z i N Xd ( i ) . Replacing Eq. (S3) in Eq. (1), gives: H Xd = − N Z N Z X i =1 N Z log N Z , (S4)as the argument of the sum does not depend on i , we can write: H Xd = − N Z N Z (cid:20) N Z log N Z (cid:21) = − N Z (cid:16) − log N Z (cid:17) = 1 . (S5) S3.2 Mobility diversity from sampled data
In sociodemographic surveys from sampled populations, individual responses are normally associated with an ex-pansion factor, a weight that accounts for the representativeness of each surveyed unit (e.g. individual, household)relative to the universe (i.e., the entirety of a population). However, the expansion factors are used to expand sam-pled populations matching them to a set of sociodemographic indicators. In our case, our main variable of interestis not sociodemographic but rather an information-theoretic one: the mobility diversity.However, for the city of Medell´ın, the data for the year of 2017 does not come with the expansion factors.To ensure the validity of our cross-years comparisons for the
MDE data, it is crucial that we assess whether themobility diversity, when computed from the non-expanded data, can still support similar qualitative conclusions.Thus, for the datasets that contained the expansion factors, we show the mobility diversity distributions obtainedfrom the unweighted samples with the ones produced by the expanded samples.We show in Figure S5 the distributions of the mobility diversity of travels made with work , nonwork or all purposes for the regions of MDE , BGT and
SAO . Comparing Figure S5 with the results of Figure 3 (using theexpansion factors), we identify that most of our main findings are observed for both samples. First, there is anincrease in mobility diversity for work travels in
MDE and
BGT , and there is a decrease in mobility diversity in
SAO from 1997 to 2007. Second, work travels distributions show smaller values of H than all and nonwork travels. Third, the nonwork purpose of travels also plays a role in the spatial distribution of travels.The major difference between the results of our data using expansion factor (Figure 3) or not (Figure S5) isthe magnitude of the mobility diversity differences between the travel types. We observe that the comparison of work travels with nonwork and all travels are not completely captured by the data sample without expansionfactor. Nonetheless, the conclusions from the data using or not the expansion factor still hold in general the same.However, we identify differences in the relationship between groups, so we highlight that the use of the expansionfactor is profoundly important to convey the right comparisons between groups. Thus, our analyses focus on thedata sample using the expansion factor, and when it is necessary, we highlight differences between the results usingthe expansion factor or not for the case of Medell´ın in 2017. Year . . . . . H (a) MDE
Year . . . . . . . . (b) BGT
Year . . . . . . . . . (c) SAO
Travel type allnonworkwork
Supplementary Figure S5: Distribution of the bootstrapped mobility diversity, H , for all , work , and nonwork travels made within each region and year. The travels are directly obtained from the surveys, without consideringthe expansion factors. 7 gender This section explores the role of gender in mobility diversity by studying the overall H distribution and the traveldiversity of men and women. We focus our attention on the work-related travels ( work ) in comparison with thediversity produced by the trips of all the travel purposes. The results of this section are aligned with Figure 5.For the case of MDE (Figure S6), we observe that men exhibit higher values of H than women. The valuesof the peak-to-peak distances between the KDEs of H corresponding to all and men travels are smaller thanthe all and women pair, a consequence of the fact that men account for the majority of the trips in the datasets(see Table S2). We also observe that the distance between the distributions of women and men decreases over theyears.Despite the fact that in MDE
BGT and even
SAO . Indeed, based onFigures 3 and S7, there is evidence that mobility diversity increases over the years. .
86 0 .
88 0 .
90 0 . a ll t r a v e l s K D E ( H ) (a) A M WAMW −− − . − . . .
86 0 .
88 0 .
90 0 . (b) A M WAMW −− − . − . . .
86 0 .
88 0 .
90 0 . H w o r k t r a v e l s K D E ( H ) (c) A M WAMW −− . − . . .
86 0 .
88 0 .
90 0 . H (d) A M WAMW −− − . − . . All (A)
Men (M)
Women (W)
Supplementary Figure S6: Kernel Density Estimation plots of the mobility diversity, H , for all travels (panels a,b ) and work travels (panels c,d ) in MDE . For each travel purpose, we plot the
KDE( H ) for travels made bymen ( M ), women ( W ), and all ( A ) travellers. The matrix appearing in the top left corner of each panel reports thepeak-to-peak distance (i.e. the distance between the median of the distributions) multiplied by a factor of − .The KDE s are computed from a distribution of H obtained by bootstrapping 1000 times 80% of the availabletravel records. 8
005 2017
Year . . . . . . . H a ll t r a v e l s (a)Sample (MDE) Year . . . . . . . (b)Exp.Fac. (MDE) Year . . . . . . . H w o r k t r a v e l s (c)Sample (MDE) Year . . . . . . . (d)Exp.Fac. (MDE)MenWomen Supplementary Figure S7: Comparing the distributions of the mobility diversity ( H ) for all travels (panels a,b )and work travels (panels c,d ) within the MDE area . Panels a and c display the case of raw travel records, whereaspanels b and d display the case of travel records obtained using the expansion factors for year 2005.In the case of SAO (Figure S8) we observe that, in general, the value of H associated with men’s mobility tendto be higher than women’s, regardless of the travel’s purpose.9 .
91 0 .
92 0 .
93 0 .
94 0 . a ll t r a v e l s K D E ( H ) (a) A M WAMW −− . − . . .
91 0 .
92 0 .
93 0 .
94 0 . (b) A M WAMW −− . − − . − . .
91 0 .
92 0 .
93 0 .
94 0 . (c) A M WAMW −− − . − . . .
91 0 .
92 0 .
93 0 .
94 0 . H w o r k t r a v e l s K D E ( H ) (d) A M WAMW −− . − . − . .
91 0 .
92 0 .
93 0 .
94 0 . H (e) A M WAMW −− . − − . − . .
91 0 .
92 0 .
93 0 .
94 0 . H (f) A M WAMW −− − . − . . All (A)
Men (M)
Women (W)
Supplementary Figure S8: Distributions of the mobility diversity, H , for all travels (panels a , b , and c ) and work travels (panels d , e , and f ) made in SAO . For each travel purpose, we plot the
KDE( H ) for travels made bymen ( M ), women ( W ), and all travellers ( A ). The matrix appearing in the top left corner of each panel reports thepeak-to-peak distance (i.e. the distance between the median of the distributions) multiplied by a factor of − .The KDE s are computed from a distribution of H obtained by bootstrapping 1000 times 80% of the availablerecords. 10ne hallmark of the gender-centred differences in urban mobility is that, on average, women are more likelyto perform shorter travels than men. This pattern can also be observed in our data as shown in Figure S9, andTables S4 and S5; The travel distance is computed by the distance between the centroids of two zones. However,such a difference in the travel distances distribution l does not hinder the chance that travellers (either women ormen) display small values of H . The reason is that, in principle, men could have longer travel distance whileconcentrating their travels in a small number of zones, which is not the case. In this way, the fact that women aremore likely to have a shorter travel distance than men would not necessarily impact the mobility diversity of thetravels performed by women.On the other hand, the fact that women are more likely to stay in the same zone could impact mobility diversitybecause they are less likely to endeavour to other zones. Women and men display a similar fraction of travelsregardless of the travel’s purpose (all or to work), and the fact that the origin and destination zones are the same(travels inside a zone) (see Table S6). The fraction of travels that travellers live and work in the same zone aresimilar for women and men (see Table S6). Moreover, women and men in general work in only one zone (seeFigure S10), but men are slightly more likely to work in more than one zone. Thus, we argue that travels insidezones and the number of workplaces are not impacting differences in the mobility diversity of women and men.Then, we check whether the majority of the zones are more likely to be visited by men than women for differentpurposes of travel. Table S7 shows the percentage of zones for which there are more travels performed by menthan by women for all , work and nonwork purposes. For all travels, MDE and
SAO show a majority of areasbeing visited by men, and
BGT shows a majority of areas being visited by women. For work travels, regardless ofthe region, the majority of the areas are mostly visited by men, and the opposite happens to nonwork travels. Weconclude that women are more likely to be concentrated in a small number of areas to work, and they are also theminority in the majority of the areas.Supplementary Figure S9: Complementary cumulative probability distribution function, P > ( l ) , of the probabilityof making a travel with a distance between origin and destination zones equal to or greater than l . Each panelrefers to a different metropolitan area.Supplementary Table S4: Minimum ( min( l ) ), maximum ( max( l ) ), median ( med ( l ) ), average ( h l i ), and standarderror of the mean ( ε l ) of the travel distance l (measured in m) made by men and women in each region and year. City Year Gender min( l ) max( l ) med ( l ) h l i ε l MDE men women men women
BGT men women men women
SAO men women men women men women p -values of the Kolmogorov–Smirnov ( KST est ) and Student t ( T T est ) tests com-paring the travel distance performed by men (M), women (W) and all travellers (A). The symbol ∗∗∗ representsthat the p -value is smaller than 0.001. City Year
KST est ( M W ) KST est ( AM ) KST est ( AW ) T T est ( M W ) T T est ( AM ) T T est ( AW ) MDE ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗
BGT ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗
SAO ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗
Supplementary Table S6: Percentages of the travels for which the origin and destination zones, P X , are the same:the percentages of travels performed by men (M) and women (W) in all travels ( P Xall ), the percentages of worktravels performed by men (M) and women (W) ( P Xwork ), and the percentages of work travels performed by men (M) and women (W) at the same zone that travellers live ( P Xlive = work ). City Year P Mall (%) P Wall (%) P Mwork (%) P Wwork (%) P Mlive = work (%) P Wlive = work (%) MDE
BGT
SAO
Supplementary Figure S10: Fraction of the number of locations in which an individual works, f n work , disaggre-gated according to the gender of the travellers.Supplementary Table S7: Percentage of areas for which the fraction of travels performed by men is higher than thethe same quantity computed for women for all , work and nonwork travels ( P Mall,area > P
Wall,area , P Mwork,area >P Wwork,area , P Mnonwork,area > P
Wnonwork,area ). City Year P Mall,area > P
Wall,area P Mwork,area > P
Wwork,area P Mnonwork,area > P
Wnonwork,area
MDE
BGT
SAO socioeconomic groups As done in Section 3.3 of the main manuscript, here, we show the distributions of the mobility diversity H for MDE and
SAO areas in Figures S11 and S12 respectively. In
MDE , we observe that upper-income travellers displaya lower mobility diversity, whereas middle-income populations tend to present the highest H values. . . . a ll t r a v e l s K D E ( H ) (a) A Low Mid UpALowMidUp −−− . −− . − . − . . . . . . (b) A Low Mid UpALowMidUp −−− . −− . . − . . . . . . H w o r k t r a v e l s K D E ( H ) (c) A Low Mid UpALowMidUp −−− . −− . − . − . . . . . . H (d) A Low Mid UpALowMidUp −−− . −− . . − . . . All (A)
Lower (Low)
Middle (Mid)
Upper (Up)
Supplementary Figure S11: KDE plots of the mobility diversity H for all travels (panels a and b ), and work travels (panels c and d ) in Medell´ın. The matrix in the top left corner of each graph reports the peak-to-peakdistance between the median of the distribution, multiplied by a factor of − .In SAO , the mobility diversity of lower and middle-income travellers display similar changes over the years,while the upper class shows the opposite trend. We argue here that the impact on the mobility of lower and middleclasses in 2007 could have been largely influenced by the profound economic changes Brazil underwent duringthat period [ ? , ? ]. .
75 0 .
80 0 .
85 0 .
90 0 . a ll t r a v e l s K D E ( H ) (a) A Low Mid UpALowMidUp −−− . −− . − . − . . . .
75 0 .
80 0 .
85 0 .
90 0 . (b) A Low Mid UpALowMidUp −−− . −− . − . − . − . . .
75 0 .
80 0 .
85 0 .
90 0 . (c) A Low Mid UpALowMidUp −−− . −− . − . − . . . .
75 0 .
80 0 .
85 0 .
90 0 . H w o r k t r a v e l s K D E ( H ) (d) A Low Mid UpALowMidUp −−− − . −− . . − . . . .
75 0 .
80 0 .
85 0 .
90 0 . H (e) A Low Mid UpALowMidUp −−− . −− . − . − . − . . .
75 0 .
80 0 .
85 0 .
90 0 . H (f) A Low Mid UpALowMidUp −−− . −− . − . − . . . All (A)
Lower (Low)
Middle (Mid)
Upper (Up)
Supplementary Figure S12: KDE plots of the mobility diversity H for all travels (panels a , b , and c ), and work travels (panels d , e , and f ) in S˜ao Paulo. The matrix in the top left corner of each graph reports the peak-to-peakdistance between the median of the distribution, multiplied by a factor of − .13 gender and socioeconomic groups In Figure S13, we display the distributions of H computed for travels made for work purposes by all combinationsof gender and socioeconomic status. As we also see in Figure 7, regardless of the purposes, the socioeconomicstatus shapes the mobility of people considerably, whereas gender exerts a smaller effect. A marked gender splitis also seen in both figures. Within each socioeconomic group, men consistently display higher values of H .On average, the gender-centred differences within each socioeconomic group tend to decrease over time, sug-gesting that a possible gender-level difference in mobility is in fact reducing (Table S8). H (a) MDE (b)
BGT (c)
SAO
Work travels
AllMen Lower classWomen Lower classMen Middle classWomen Middle classMen Upper classWomen Upper class
Supplementary Figure S13: Distribution of the mobility diversity, H , for travels made by work purposes bytravellers grouped according to their socioeconomic status and gender. Each column refers to a different region,and for each region, we consider all the available years. For each socioeconomic status ( upper , middle , and lower ) darker hue denotes men travellers, whereas lighter hue denotes women ones.14upplementary Table S8: Gender differences, median ( H MS ) − median ( H WS ) , of the mobility diversity H oftravels made for all and work purposes by travellers grouped according to their socioeconomic status, S , andgender, X = M, W . The values report the peak-to-peak distance between the median of the distribution of H ,multiplied by a factor of − . The values in bold represent the cases in which median ( H WS ) > median ( H MS ) . City Year Purpose lower middle upperMDE all work all work
BGT all work all -1.8 work -0.6
SAO all work all -0.2 -8.0 work -7.3 all work
S4 Statistical verification of the mobility diversity distributions
As described in the main manuscript, to account for variations in sample sizes, we employed a bootstrappingstrategy to estimate mobility diversity distribution. From these distributions, we used different statistical methodsto verify the differences in the diversity distributions across groups. The tests we used were the Welch’s t -test [ ? ],the ANOVA [ ? ], and the Tukey’s HSD post hoc test [ ? ]. The Welch’s t -test compares if the distributions ofmobility diversity are statistically different from each other. Secondly, the ANOVA test compares if the averagesof the groups’ mobility diversity distributions are statistically different, expressing if the result extracted for eachelement in a group is in fact, different from the other elements. Finally, the Tukey’s HSD post hoc test indicateswhat pairs of groups’ means are different. The statistical tests were computed using the following Python packages:pandas, numpy, scipy, statsmodels, and pingouin.To evaluate the contribution of the gender and socioeconomic dimensions to the mobility diversity H , we firstapply the ANOVA one-way and two-way tests to identify whether the distributions present similar average valuesof H . In Tables S9 and S10, we plot all the values of F -statistic and p -value of the ANOVA test computed fromthe mobility diversity H of travels made for all and work purposes by travellers aggregated by gender andsocioeconomic status.First, we test if the values of H from the gender groups are from populations with the same mean values.Considering the all and work travels, we can reject the null hypothesis that the mean values of H from men,women and all travellers are statistically the same because the p -values are smaller than 0.01 and the F -valuesare not low. Next, as the ANOVA test does not specify which specific groups differ from each other, we applythe Tukey’s HSD post hoc test to discover whether the specific groups hold mutually statistically different. Forinstance, Tukey’s test can tell us if the mean values of H computed for men travellers are not statistically differentfrom the same quantities computed for all travellers but, instead, are statistically different from the women’scounterparts.Applying Tukey’s HSD post hoc test, see Tables S11 - S17, we observe that the p -values from the multi-groupmeans comparisons of the women , men and all distributions of the mobility diversity between the differentpurpose of travels are smaller than 0.01. The same procedure using ANOVA and Tukey’s HSD post hoc testsis applied for the values of H obtained when grouping travellers according to their socioeconomic classes. The p -values of the mobility diversity calculated from all and work travels using the ANOVA test are all smaller15han 0.01, and the F -values are even higher than their counterpart computed for the gender-based classification.Using the Tukey’s HSD post hoc tests, we can reject that the values of H of the socioeconomic groups are frompopulations with same mean values.The two-way ANOVA test is then used to analyse the relationship between gender and socioeconomic statusin the measurement of mobility diversity. The values of F seem to be low, so the ANOVA results do not excludethat the mobility diversity, H , of travellers belonging to different socioeconomic and gender groups belong to thesame distribution. To exclude such possibility, we decided to apply the Tukey’s HSD test. From all the multi-groupmeans comparisons of the distributions of the mobility diversity between different set of travels, see Tables S11 -S17, we can not reject that the values of H of the following distributions are from populations with same meanvalues: men from lower and middle classes of MDE in 2005, and men and women from middle class of
SAO in 1997.In Section 3, we observe that we can not reject that the distributions of the work travels are statistically similar(tested by Welch’s t -test with p -value < . ) in three cases: (i) comparing women and men from the upper classof BGT in 2019; (ii) comparing women and men from the middle class of
SAO in 1997; and (iii) comparing menfrom the lower and middle class of
MDE in 2005. These results of Section 3 are in agreement with the ones ofthe Tukey’s test.In general, we observe that the null hypotheses of gender groups and socioeconomic groups displaying similarmean values of mobility diversity can be rejected. When gender and socioeconomic dimensions are combined,for the majority of the cases, we can reject that the distributions are obtained from populations with the samemean values. Thus, each gender and socioeconomic group alone or taken together display different distributionsof mobility diversity. Such a difference in the distributions of H means that gender and socioeconomic differencesare based on how each group explores the space available.Supplementary Table S9: F Statistic of the ANOVA Test computed from the mobility diversity of all travels. Allthe p -values are smaller than 0.001. Location
MDE BGT SAO
Years 2005 2017 2012 2019 1997 2007 2017Gender Groups 10030 14650 95412 4083 5159 29837 9989Socioeconomic Groups 4.54e+06 1.08e+07 4.24e+06 666131 1.21e+06 1.41e+06 2.42e+06Combined Groups 93.30 52.52 3.16 61.58 227.55 817.55 13.64
Supplementary Table S10: F Statistic ( p -value) of the ANOVA Test computed from the mobility diversity of worktravels. All the p -values are smaller than 0.001. Location
MDE BGT SAO
Years 2005 2017 2012 2019 1997 2007 2017Gender Groups 8239 8084 23135 5833 9150 7965 14399Socioeconomic Groups 170023 1.94e+06 450620 335207 402173 308007 736732Combined Groups 55.92 44.08 102.71 25.72 652.40 382.06 118.68
MDE ∗∗∗ symbol denotes a p -value smaller than 0.001.We highlight the cells of groups having p -values higher than 0.001. Travels Groups Mean 95% Confidence interval Adjusteddifference Lower bound Upper bound p -value all ( all ) × ( men ) 0.0001 0.0001 0.0002 ∗∗∗ ( all ) × ( women ) -0.0031 -0.0032 -0.0031( men ) × ( women ) -0.0033 -0.0034 -0.0032( all ) × ( lower ) -0.0923 -0.0925 -0.0921( all ) × ( middle ) -0.0375 -0.0377 -0.0373( all ) × ( upper ) -0.1877 -0.1878 -0.1875( lower ) × ( middle ) 0.0548 0.0546 0.055( lower ) × ( upper ) -0.0953 -0.0955 -0.0951( middle ) × ( upper ) -0.1501 -0.1503 -0.1499( all ) × ( men - lower ) -0.0883 -0.0886 -0.088( all ) × ( men - middle ) -0.0382 -0.0385 -0.0378( all ) × ( men - upper ) -0.185 -0.1853 -0.1847( all ) × ( women - lower ) -0.1013 -0.1016 -0.101( all ) × ( women - middle ) -0.0415 -0.0418 -0.0411( all ) × ( women - upper ) -0.2114 -0.2117 -0.211( men - lower ) × ( men - middle ) 0.0501 0.0498 0.0504( men - lower ) × ( men - upper ) -0.0967 -0.097 -0.0964( men - lower ) × ( women - lower ) -0.013 -0.0133 -0.0127( men - lower ) × ( women - middle ) 0.0468 0.0465 0.0471( men - lower ) × ( women - upper ) -0.1231 -0.1234 -0.1228( men - middle ) × ( men - upper ) -0.1468 -0.1471 -0.1465( men - middle ) × ( women - lower ) -0.0631 -0.0635 -0.0628( men - middle ) × ( women - middle ) -0.0033 -0.0036 -0.003( men - middle ) × ( women - upper ) -0.1732 -0.1735 -0.1729( men - upper ) × ( women - lower ) 0.0837 0.0833 0.084( men - upper ) × ( women - middle ) 0.1435 0.1432 0.1438( men - upper ) × ( women - upper ) -0.0264 -0.0267 -0.0261( women - lower ) × ( women - middle ) 0.0598 0.0595 0.0602( women - lower ) × ( women - upper ) -0.1101 -0.1104 -0.1097( women - middle ) × ( women - upper ) -0.1699 -0.1702 -0.1696 work ( all ) × ( men ) -0.004 -0.0042 -0.0038 ∗∗∗ ( all ) × ( women ) -0.0093 -0.0095 -0.0091( men ) × ( women ) -0.0053 -0.0055 -0.0051( all ) × ( lower ) -0.0149 -0.0153 -0.0144( all ) × ( middle ) -0.0097 -0.0101 -0.0092( all ) × ( upper ) -0.1251 -0.1255 -0.1246( lower ) × ( middle ) 0.0052 0.0048 0.0056( lower ) × ( upper ) -0.1102 -0.1106 -0.1098( middle ) × ( upper ) -0.1154 -0.1158 -0.115( all ) × ( men - lower ) -0.0188 -0.0195 -0.0181( all ) × ( men - middle ) -0.0188 -0.0195 -0.0181( all ) × ( men - upper ) -0.1568 -0.1575 -0.1561( all ) × ( women - lower ) -0.0398 -0.0405 -0.0391( all ) × ( women - middle ) -0.022 -0.0226 -0.0213( all ) × ( women - upper ) -0.1682 -0.1689 -0.1675( men - lower ) × ( men - middle ) 0.0 -0.0007 0.0007 0.9( men - lower ) × ( men - upper ) -0.138 -0.1387 -0.1373 ∗∗∗ ( men - lower ) × ( women - lower ) -0.021 -0.0217 -0.0203( men - lower ) × ( women - middle ) -0.0032 -0.0039 -0.0025( men - lower ) × ( women - upper ) -0.1494 -0.1501 -0.1487( men - middle ) × ( men - upper ) -0.138 -0.1387 -0.1373( men - middle ) × ( women - lower ) -0.021 -0.0217 -0.0203( men - middle ) × ( women - middle ) -0.0032 -0.0039 -0.0025( men - middle ) × ( women - upper ) -0.1494 -0.1501 -0.1487( men - upper ) × ( women - lower ) 0.117 0.1163 0.1177( men - upper ) × ( women - middle ) 0.1348 0.1341 0.1355( men - upper ) × ( women - upper ) -0.0114 -0.0121 -0.0107( women - lower ) × ( women - middle ) 0.0178 0.0171 0.0185( women - lower ) × ( women - upper ) -0.1284 -0.1291 -0.1277( women - middle ) × ( women - upper ) -0.1462 -0.1469 -0.1455 MDE
Travels Groups Mean 95% Confidence interval Adjusteddifference Lower bound Upper bound p -value all ( all ) × ( men ) 0.0009 0.0008 0.0009 ∗∗∗ ( all ) × ( women ) -0.0018 -0.0018 -0.0017( men ) × ( women ) -0.0027 -0.0027 -0.0026( all ) × ( lower ) -0.0065 -0.0066 -0.0065( all ) × ( middle ) -0.0436 -0.0437 -0.0435( all ) × ( upper ) -0.1675 -0.1676 -0.1674( lower ) × ( middle ) -0.0371 -0.0372 -0.037( lower ) × ( upper ) -0.1609 -0.161 -0.1609( middle ) × ( upper ) -0.1239 -0.1239 -0.1238( all ) × ( men lower ) -0.0071 -0.0072 -0.007( all ) × ( men - middle ) -0.0419 -0.042 -0.0417( all ) × ( men - upper ) -0.1606 -0.1607 -0.1604( all ) × ( women lower ) -0.0075 -0.0076 -0.0074( all ) × ( women - middle ) -0.0473 -0.0474 -0.0472( all ) × ( women - upper ) -0.1788 -0.1789 -0.1787( men lower ) × ( men - middle ) -0.0348 -0.0349 -0.0346( men lower ) × ( men - upper ) -0.1535 -0.1536 -0.1533( men lower ) × ( women lower ) -0.0004 -0.0006 -0.0003( men lower ) × ( women - middle ) -0.0402 -0.0403 -0.0401( men lower ) × ( women - upper ) -0.1717 -0.1718 -0.1716( men - middle ) × ( men - upper ) -0.1187 -0.1188 -0.1186( men - middle ) × ( women lower ) 0.0344 0.0342 0.0345( men - middle ) × ( women - middle ) -0.0054 -0.0055 -0.0053( men - middle ) × ( women - upper ) -0.1369 -0.137 -0.1368( men - upper ) × ( women lower ) 0.1531 0.1529 0.1532( men - upper ) × ( women - middle ) 0.1133 0.1132 0.1134( men - upper ) × ( women - upper ) -0.0182 -0.0183 -0.0181( women lower ) × ( women - middle ) -0.0398 -0.0399 -0.0396( women lower ) × ( women - upper ) -0.1713 -0.1714 -0.1711( women - middle ) × ( women - upper ) -0.1315 -0.1316 -0.1314 work ( all ) × ( men ) 0.0003 0.0002 0.0005 ∗∗∗ ( all ) × ( women ) -0.0048 -0.0049 -0.0046( men ) × ( women ) -0.0051 -0.0052 -0.005( all ) × ( lower ) -0.0083 -0.0085 -0.0081( all ) × ( middle ) -0.0441 -0.0443 -0.0439( all ) × ( upper ) -0.1698 -0.17 -0.1696( lower ) × ( middle ) -0.0358 -0.036 -0.0357( lower ) × ( upper ) -0.1615 -0.1617 -0.1613( middle ) × ( upper ) -0.1257 -0.1259 -0.1255( all ) × ( men lower ) -0.0111 -0.0114 -0.0108( all ) × ( men - middle ) -0.0439 -0.0442 -0.0436( all ) × ( men - upper ) -0.1683 -0.1687 -0.168( all ) × ( women lower ) -0.0121 -0.0124 -0.0117( all ) × ( women - middle ) -0.0523 -0.0526 -0.0519( all ) × ( women - upper ) -0.1865 -0.1869 -0.1862( men lower ) × ( men - middle ) -0.0328 -0.0331 -0.0325( men lower ) × ( men - upper ) -0.1572 -0.1576 -0.1569( men lower ) × ( women lower ) -0.001 -0.0013 -0.0006( men lower ) × ( women - middle ) -0.0412 -0.0415 -0.0408( men lower ) × ( women - upper ) -0.1754 -0.1758 -0.1751( men - middle ) × ( men - upper ) -0.1245 -0.1248 -0.1241( men - middle ) × ( women lower ) 0.0318 0.0315 0.0322( men - middle ) × ( women - middle ) -0.0084 -0.0087 -0.0081( men - middle ) × ( women - upper ) -0.1426 -0.143 -0.1423( men - upper ) × ( women lower ) 0.1563 0.1559 0.1566( men - upper ) × ( women - middle ) 0.1161 0.1157 0.1164( men - upper ) × ( women - upper ) -0.0182 -0.0185 -0.0178( women lower ) × ( women - middle ) -0.0402 -0.0406 -0.0399( women lower ) × ( women - upper ) -0.1745 -0.1748 -0.1741( women - middle ) × ( women - upper ) -0.1342 -0.1346 -0.1339 BGT
Travels Groups Mean 95% Confidence interval Adjusteddifference Lower bound Upper bound p -value all ( all ) × ( men ) 0.0052 0.0051 0.0053 ∗∗∗ ( all ) × ( women ) -0.0125 -0.0126 -0.0124( men ) × ( women ) -0.0177 -0.0178 -0.0175( all ) × ( lower ) -0.1152 -0.1154 -0.115( all ) × ( middle ) -0.025 -0.0252 -0.0247( all ) × ( upper ) -0.2214 -0.2216 -0.2212( lower ) × ( middle ) 0.0903 0.0901 0.0905( lower ) × ( upper ) -0.1061 -0.1063 -0.1059( middle ) × ( upper ) -0.1964 -0.1966 -0.1962( all ) × ( men lower ) -0.1136 -0.1139 -0.1132( all ) × ( men - middle ) -0.0217 -0.0221 -0.0214( all ) × ( men - upper ) -0.2239 -0.2243 -0.2236( all ) × ( women lower ) -0.1282 -0.1286 -0.1279( all ) × ( women - middle ) -0.0407 -0.0411 -0.0404( all ) × ( women - upper ) -0.2435 -0.2439 -0.2432( men lower ) × ( men - middle ) 0.0918 0.0915 0.0922( men lower ) × ( men - upper ) -0.1104 -0.1107 -0.11( men lower ) × ( women lower ) -0.0146 -0.015 -0.0143( men lower ) × ( women - middle ) 0.0728 0.0725 0.0732( men lower ) × ( women - upper ) -0.13 -0.1303 -0.1296( men - middle ) × ( men - upper ) -0.2022 -0.2025 -0.2018( men - middle ) × ( women lower ) -0.1065 -0.1068 -0.1061( men - middle ) × ( women - middle ) -0.019 -0.0193 -0.0186( men - middle ) × ( women - upper ) -0.2218 -0.2221 -0.2214( men - upper ) × ( women lower ) 0.0957 0.0954 0.0961( men - upper ) × ( women - middle ) 0.1832 0.1828 0.1835( men - upper ) × ( women - upper ) -0.0196 -0.02 -0.0193( women lower ) × ( women - middle ) 0.0875 0.0871 0.0878( women lower ) × ( women - upper ) -0.1153 -0.1157 -0.115( women - middle ) × ( women - upper ) -0.2028 -0.2032 -0.2025 work ( all ) × ( men ) -0.0117 -0.0119 -0.0114 ∗∗∗ ( all ) × ( women ) -0.0166 -0.0168 -0.0163( men ) × ( women ) -0.0049 -0.0052 -0.0047( all ) × ( lower ) -0.0139 -0.0143 -0.0135( all ) × ( middle ) -0.0295 -0.0299 -0.0291( all ) × ( upper ) -0.1689 -0.1693 -0.1685( lower ) × ( middle ) -0.0156 -0.016 -0.0152( lower ) × ( upper ) -0.1549 -0.1553 -0.1545( middle ) × ( upper ) -0.1394 -0.1397 -0.139( all ) × ( men lower ) -0.0326 -0.0332 -0.032( all ) × ( men - middle ) -0.0498 -0.0504 -0.0492( all ) × ( men - upper ) -0.195 -0.1956 -0.1944( all ) × ( women lower ) -0.0488 -0.0494 -0.0482( all ) × ( women - middle ) -0.0576 -0.0582 -0.0569( all ) × ( women - upper ) -0.2311 -0.2317 -0.2305( men lower ) × ( men - middle ) -0.0172 -0.0178 -0.0166( men lower ) × ( men - upper ) -0.1624 -0.163 -0.1618( men lower ) × ( women lower ) -0.0162 -0.0168 -0.0156( men lower ) × ( women - middle ) -0.025 -0.0256 -0.0244( men lower ) × ( women - upper ) -0.1985 -0.1991 -0.1979( men - middle ) × ( men - upper ) -0.1452 -0.1458 -0.1446( men - middle ) × ( women lower ) 0.001 0.0004 0.0016( men - middle ) × ( women - middle ) -0.0078 -0.0084 -0.0072( men - middle ) × ( women - upper ) -0.1813 -0.1819 -0.1807( men - upper ) × ( women lower ) 0.1462 0.1456 0.1468( men - upper ) × ( women - middle ) 0.1374 0.1368 0.138( men - upper ) × ( women - upper ) -0.0361 -0.0367 -0.0355( women lower ) × ( women - middle ) -0.0088 -0.0094 -0.0082( women lower ) × ( women - upper ) -0.1823 -0.1829 -0.1817( women - middle ) × ( women - upper ) -0.1735 -0.1741 -0.1729 BGT
Travels Groups Mean 95% Confidence interval Adjusteddifference Lower bound Upper bound p -value all ( all ) × ( men ) 0.0002 0.0001 0.0002 ∗∗∗ ( all ) × ( women ) -0.001 -0.001 -0.0009( men ) × ( women ) -0.0012 -0.0012 -0.0011( all ) × ( lower ) -0.0102 -0.0103 -0.0102( all ) × ( middle ) -0.0028 -0.0028 -0.0027( all ) × ( upper ) -0.0398 -0.0399 -0.0397( lower ) × ( middle ) 0.0075 0.0074 0.0076( lower ) × ( upper ) -0.0295 -0.0296 -0.0295( middle ) × ( upper ) -0.037 -0.0371 -0.037( all ) × ( men lower ) -0.0094 -0.0096 -0.0093( all ) × ( men - middle ) -0.0034 -0.0036 -0.0033( all ) × ( men - upper ) -0.0463 -0.0464 -0.0461( all ) × ( women lower ) -0.0128 -0.0129 -0.0126( all ) × ( women - middle ) -0.0036 -0.0037 -0.0035( all ) × ( women - upper ) -0.0445 -0.0446 -0.0443( men lower ) × ( men - middle ) 0.006 0.0059 0.0061( men lower ) × ( men - upper ) -0.0368 -0.037 -0.0367( men lower ) × ( women lower ) -0.0033 -0.0035 -0.0032( men lower ) × ( women - middle ) 0.0058 0.0057 0.006( men lower ) × ( women - upper ) -0.035 -0.0352 -0.0349( men - middle ) × ( men - upper ) -0.0428 -0.043 -0.0427( men - middle ) × ( women lower ) -0.0093 -0.0095 -0.0092( men - middle ) × ( women - middle ) -0.0002 -0.0003 -0.0( men - middle ) × ( women - upper ) -0.041 -0.0411 -0.0409( men - upper ) × ( women lower ) 0.0335 0.0334 0.0336( men - upper ) × ( women - middle ) 0.0427 0.0425 0.0428( men - upper ) × ( women - upper ) 0.0018 0.0017 0.0019( women lower ) × ( women - middle ) 0.0092 0.009 0.0093( women lower ) × ( women - upper ) -0.0317 -0.0318 -0.0316( women - middle ) × ( women - upper ) -0.0409 -0.041 -0.0407 work ( all ) × ( men ) -0.0015 -0.0017 -0.0014 ∗∗∗ ( all ) × ( women ) -0.004 -0.0041 -0.0039( men ) × ( women ) -0.0025 -0.0026 -0.0024( all ) × ( lower ) -0.0204 -0.0206 -0.0202( all ) × ( middle ) -0.0066 -0.0068 -0.0064( all ) × ( upper ) -0.0827 -0.0829 -0.0824( lower ) × ( middle ) 0.0138 0.0136 0.014( lower ) × ( upper ) -0.0622 -0.0625 -0.062( middle ) × ( upper ) -0.0761 -0.0763 -0.0758( all ) × ( men lower ) -0.0231 -0.0235 -0.0228( all ) × ( men - middle ) -0.0101 -0.0105 -0.0097( all ) × ( men - upper ) -0.111 -0.1114 -0.1106( all ) × ( women lower ) -0.0306 -0.0309 -0.0302( all ) × ( women - middle ) -0.0151 -0.0154 -0.0147( all ) × ( women - upper ) -0.1106 -0.1109 -0.1102( men lower ) × ( men - middle ) 0.013 0.0127 0.0134( men lower ) × ( men - upper ) -0.0879 -0.0882 -0.0875( men lower ) × ( women lower ) -0.0074 -0.0078 -0.0071( men lower ) × ( women - middle ) 0.0081 0.0077 0.0084( men lower ) × ( women - upper ) -0.0874 -0.0878 -0.0871( men - middle ) × ( men - upper ) -0.1009 -0.1013 -0.1005( men - middle ) × ( women lower ) -0.0204 -0.0208 -0.0201( men - middle ) × ( women - middle ) -0.005 -0.0053 -0.0046( men - middle ) × ( women - upper ) -0.1005 -0.1008 -0.1001( men - upper ) × ( women lower ) 0.0805 0.0801 0.0808( men - upper ) × ( women - middle ) 0.0959 0.0956 0.0963( men - upper ) × ( women - upper ) 0.0004 0.0001 0.0008( women lower ) × ( women - middle ) 0.0155 0.0151 0.0158( women lower ) × ( women - upper ) -0.08 -0.0804 -0.0797( women - middle ) × ( women - upper ) -0.0955 -0.0959 -0.0951 SAO
Travels Groups Mean 95% Confidence interval Adjusteddifference Lower bound Upper bound p -value all ( all ) × ( men ) -0.0001 -0.0002 -0.0001 ∗∗∗ ( all ) × ( women ) -0.0022 -0.0023 -0.0021( men ) × ( women ) -0.0021 -0.0022 -0.002( all ) × ( lower ) -0.0428 -0.043 -0.0427( all ) × ( middle ) -0.0041 -0.0043 -0.004( all ) × ( upper ) -0.0819 -0.082 -0.0818( lower ) × ( middle ) 0.0387 0.0386 0.0388( lower ) × ( upper ) -0.0391 -0.0392 -0.0389( middle ) × ( upper ) -0.0778 -0.0779 -0.0777( all ) × ( men lower ) -0.039 -0.0392 -0.0388( all ) × ( men - middle ) -0.0054 -0.0056 -0.0052( all ) × ( men - upper ) -0.0819 -0.0822 -0.0817( all ) × ( women lower ) -0.0527 -0.0529 -0.0525( all ) × ( women - middle ) -0.0059 -0.0062 -0.0057( all ) × ( women - upper ) -0.1014 -0.1016 -0.1012( men lower ) × ( men - middle ) 0.0336 0.0334 0.0338( men lower ) × ( men - upper ) -0.043 -0.0432 -0.0427( men lower ) × ( women lower ) -0.0137 -0.0139 -0.0135( men lower ) × ( women - middle ) 0.0331 0.0328 0.0333( men lower ) × ( women - upper ) -0.0624 -0.0627 -0.0622( men - middle ) × ( men - upper ) -0.0766 -0.0768 -0.0763( men - middle ) × ( women lower ) -0.0473 -0.0475 -0.0471( men - middle ) × ( women - middle ) -0.0006 -0.0008 -0.0003( men - middle ) × ( women - upper ) -0.096 -0.0963 -0.0958( men - upper ) × ( women lower ) 0.0293 0.029 0.0295( men - upper ) × ( women - middle ) 0.076 0.0758 0.0762( men - upper ) × ( women - upper ) -0.0195 -0.0197 -0.0192( women lower ) × ( women - middle ) 0.0468 0.0465 0.047( women lower ) × ( women - upper ) -0.0487 -0.049 -0.0485( women - middle ) × ( women - upper ) -0.0955 -0.0957 -0.0952 work ( all ) × ( men ) -0.0047 -0.0049 -0.0046 ∗∗∗ ( all ) × ( women ) -0.0041 -0.0042 -0.0039( men ) × ( women ) 0.0007 0.0005 0.0008( all ) × ( lower ) 0.0003 0.0001 0.0006( all ) × ( middle ) -0.0125 -0.0127 -0.0122( all ) × ( upper ) -0.0978 -0.098 -0.0976( lower ) × ( middle ) -0.0128 -0.0131 -0.0126( lower ) × ( upper ) -0.0981 -0.0984 -0.0979( middle ) × ( upper ) -0.0853 -0.0856 -0.0851( all ) × ( men - lower ) -0.008 -0.0085 -0.0076( all ) × ( men - middle ) -0.0183 -0.0188 -0.0179( all ) × ( men - upper ) -0.1076 -0.1081 -0.1072( all ) × ( women - lower ) -0.014 -0.0144 -0.0135( all ) × ( women - middle ) -0.0185 -0.0189 -0.018( all ) × ( women - upper ) -0.1537 -0.1541 -0.1532( men lower ) × ( men - middle ) -0.0103 -0.0107 -0.0099( men lower ) × ( men - upper ) -0.0996 -0.1 -0.0992( men lower ) × ( women - lower ) -0.0059 -0.0064 -0.0055( men lower ) × ( women - middle ) -0.0104 -0.0109 -0.01( men lower ) × ( women - upper ) -0.1456 -0.1461 -0.1452( men - middle ) × ( men - upper ) -0.0893 -0.0897 -0.0889( men - middle ) × ( women - lower ) 0.0044 0.004 0.0048( men - middle ) × ( women - middle ) -0.0001 -0.0005 0.0003 0.9( men - middle ) × ( women - upper ) -0.1353 -0.1357 -0.1349 ∗∗∗ ( men - upper ) × ( women - lower ) 0.0937 0.0933 0.0941( men - upper ) × ( women - middle ) 0.0892 0.0887 0.0896( men - upper ) × ( women - upper ) -0.046 -0.0464 -0.0456( women lower ) × ( women - middle ) -0.0045 -0.0049 -0.0041( women lower ) × ( women - upper ) -0.1397 -0.1401 -0.1393( women - middle ) × ( women - upper ) -0.1352 -0.1356 -0.1348 SAO
Travels Groups Mean 95% Confidence interval Adjusteddifference Lower bound Upper bound p -value all ( all ) × ( men ) -0.0029 -0.0029 -0.0028 ∗∗∗ ( all ) × ( women ) 0.0014 0.0014 0.0014( men ) × ( women ) 0.0043 0.0042 0.0043( all ) × ( lower ) -0.0373 -0.0374 -0.0373( all ) × ( middle ) -0.0041 -0.0042 -0.0041( all ) × ( upper ) -0.0283 -0.0283 -0.0282( lower ) × ( middle ) 0.0332 0.0331 0.0333( lower ) × ( upper ) 0.0091 0.009 0.0091( middle ) × ( upper ) -0.0241 -0.0242 -0.0241( all ) × ( men - lower ) -0.0395 -0.0396 -0.0393( all ) × ( men - middle ) -0.0092 -0.0093 -0.0091( all ) × ( men - upper ) -0.0299 -0.03 -0.0297( all ) × ( women - lower ) -0.0392 -0.0394 -0.0391( all ) × ( women - middle ) -0.0011 -0.0013 -0.001( all ) × ( women - upper ) -0.0368 -0.0369 -0.0366( men lower ) × ( men - middle ) 0.0303 0.0302 0.0304( men lower ) × ( men - upper ) 0.0096 0.0095 0.0097( men lower ) × ( women - lower ) 0.0002 0.0001 0.0003( men lower ) × ( women - middle ) 0.0383 0.0382 0.0384( men lower ) × ( women - upper ) 0.0027 0.0026 0.0028( men - middle ) × ( men - upper ) -0.0207 -0.0208 -0.0206( men - middle ) × ( women - lower ) -0.0301 -0.0302 -0.03( men - middle ) × ( women - middle ) 0.008 0.0079 0.0081( men - middle ) × ( women - upper ) -0.0276 -0.0277 -0.0275( men - upper ) × ( women - lower ) -0.0094 -0.0095 -0.0093( men - upper ) × ( women - middle ) 0.0287 0.0286 0.0288( men - upper ) × ( women - upper ) -0.0069 -0.007 -0.0068( women lower ) × ( women - middle ) 0.0381 0.038 0.0382( women lower ) × ( women - upper ) 0.0025 0.0024 0.0026( women - middle ) × ( women - upper ) -0.0356 -0.0357 -0.0355work ( all ) × ( men ) -0.0035 -0.0036 -0.0034 ∗∗∗ ( all ) × ( women ) 0.0006 0.0005 0.0007( men ) × ( women ) 0.0041 0.004 0.0042( all ) × ( lower ) -0.036 -0.0361 -0.0358( all ) × ( middle ) -0.0053 -0.0054 -0.0051( all ) × ( upper ) -0.0315 -0.0317 -0.0314( lower ) × ( middle ) 0.0307 0.0305 0.0308( lower ) × ( upper ) 0.0044 0.0043 0.0046( middle ) × ( upper ) -0.0262 -0.0264 -0.0261( all ) × ( men - lower ) -0.0415 -0.0417 -0.0412( all ) × ( men - middle ) -0.0106 -0.0109 -0.0104( all ) × ( men - upper ) -0.0393 -0.0395 -0.039( all ) × ( women - lower ) -0.0423 -0.0425 -0.042( all ) × ( women - middle ) -0.0033 -0.0035 -0.003( all ) × ( women - upper ) -0.043 -0.0433 -0.0428( men lower ) × ( men - middle ) 0.0309 0.0306 0.0311( men lower ) × ( men - upper ) 0.0022 0.0019 0.0024( men lower ) × ( women - lower ) -0.0008 -0.001 -0.0005( men lower ) × ( women - middle ) 0.0382 0.0379 0.0384( men lower ) × ( women - upper ) -0.0015 -0.0018 -0.0013( men - middle ) × ( men - upper ) -0.0287 -0.0289 -0.0284( men - middle ) × ( women - lower ) -0.0316 -0.0319 -0.0314( men - middle ) × ( women - middle ) 0.0073 0.0071 0.0076( men - middle ) × ( women - upper ) -0.0324 -0.0326 -0.0321( men - upper ) × ( women - lower ) -0.003 -0.0032 -0.0027( men - upper ) × ( women - middle ) 0.036 0.0357 0.0362( men - upper ) × ( women - upper ) -0.0037 -0.004 -0.0035( women lower ) × ( women - middle ) 0.039 0.0387 0.0392( women lower ) × ( women - upper ) -0.0008 -0.001 -0.0005( women - middle ) × ( women - upper ) -0.0397 -0.04 -0.0395 SAO
Travels Groups Mean 95% Confidence interval Adjusteddifference Lower bound Upper bound p -value all ( all ) × ( men ) 0.0002 0.0002 0.0002 ∗∗∗ ( all ) × ( women ) -0.0016 -0.0017 -0.0016( men ) × ( women ) -0.0018 -0.0018 -0.0018( all ) × ( lower ) -0.0297 -0.0297 -0.0296( all ) × ( middle ) -0.0036 -0.0037 -0.0035( all ) × ( upper ) -0.0579 -0.058 -0.0578( lower ) × ( middle ) 0.0261 0.026 0.0261( lower ) × ( upper ) -0.0282 -0.0283 -0.0282( middle ) × ( upper ) -0.0543 -0.0544 -0.0542( all ) × ( men - lower ) -0.0316 -0.0317 -0.0315( all ) × ( men - middle ) -0.0043 -0.0044 -0.0042( all ) × ( men - upper ) -0.0615 -0.0616 -0.0614( all ) × ( women - lower ) -0.0344 -0.0345 -0.0343( all ) × ( women - middle ) -0.0052 -0.0053 -0.005( all ) × ( women - upper ) -0.0656 -0.0657 -0.0655( men lower ) × ( men - middle ) 0.0274 0.0272 0.0275( men lower ) × ( men - upper ) -0.0299 -0.03 -0.0297( men lower ) × ( women - lower ) -0.0028 -0.0029 -0.0027( men lower ) × ( women - middle ) 0.0265 0.0264 0.0266( men lower ) × ( women - upper ) -0.034 -0.0341 -0.0338( men - middle ) × ( men - upper ) -0.0572 -0.0573 -0.0571( men - middle ) × ( women lower ) -0.0301 -0.0303 -0.03( men - middle ) × ( women - middle ) -0.0009 -0.001 -0.0008( men - middle ) × ( women - upper ) -0.0613 -0.0614 -0.0612( men - upper ) × ( women lower ) 0.0271 0.0269 0.0272( men - upper ) × ( women - middle ) 0.0563 0.0562 0.0564( men - upper ) × ( women - upper ) -0.0041 -0.0042 -0.004( women lower ) × ( women - middle ) 0.0293 0.0291 0.0294( women lower ) × ( women - upper ) -0.0312 -0.0313 -0.0311( women - middle ) × ( women - upper ) -0.0604 -0.0605 -0.0603 work ( all ) × ( men ) 0.0002 0.0002 0.0002 ∗∗∗ ( all ) × ( women ) -0.0016 -0.0017 -0.0016( men ) × ( women ) -0.0018 -0.0018 -0.0018( all ) × ( lower ) -0.0297 -0.0297 -0.0296( all ) × ( middle ) -0.0036 -0.0037 -0.0035( all ) × ( upper ) -0.0579 -0.058 -0.0578( lower ) × ( middle ) 0.0261 0.026 0.0261( lower ) × ( upper ) -0.0282 -0.0283 -0.0282( middle ) × ( upper ) -0.0543 -0.0544 -0.0542( all ) × ( men lower ) -0.0316 -0.0317 -0.0315( all ) × ( men - middle ) -0.0043 -0.0044 -0.0042( all ) × ( men - upper ) -0.0615 -0.0616 -0.0614( all ) × ( women lower ) -0.0344 -0.0345 -0.0343( all ) × ( women - middle ) -0.0052 -0.0053 -0.005( all ) × ( women - upper ) -0.0656 -0.0657 -0.0655( men lower ) × ( men - middle ) 0.0274 0.0272 0.0275( men lower ) × ( men - upper ) -0.0299 -0.03 -0.0297( men lower ) × ( women lower ) -0.0028 -0.0029 -0.0027( men lower ) × ( women - middle ) 0.0265 0.0264 0.0266( men lower ) × ( women - upper ) -0.034 -0.0341 -0.0338( men - middle ) × ( men - upper ) -0.0572 -0.0573 -0.0571( men - middle ) × ( women lower ) -0.0301 -0.0303 -0.03( men - middle ) × ( women - middle ) -0.0009 -0.001 -0.0008( men - middle ) × ( women - upper ) -0.0613 -0.0614 -0.0612( men - upper ) × ( women lower ) 0.0271 0.0269 0.0272( men - upper ) × ( women - middle ) 0.0563 0.0562 0.0564( men - upper ) × ( women - upper ) -0.0041 -0.0042 -0.004( women lower ) × ( women - middle ) 0.0293 0.0291 0.0294( women lower ) × ( women - upper ) -0.0312 -0.0313 -0.0311( women - middle ) × ( women - upper ) -0.0604 -0.0605 -0.0603) -0.0604 -0.0605 -0.0603