How and where people spend their time can strongly influence environmental exposures relevant to health. Time-activity patterns have been identified as one of the main sources of exposure measurement error in air pollution epidemiology [1
]. Recent studies highlighted the importance of integrating mobility for better air pollution exposure assessment [2
]. However, existing literature describing time-activity or mobility patterns and their determinants is largely from populations in urban areas of high-income countries [7
]. Little evidence is available regarding individual or contextual determinants of mobility for populations experiencing rapid economic development and urbanization. Findings from high-income countries may not be relevant to populations in low- and middle-income countries. The gap in knowledge is especially relevant for populations in peri-urban areas of India experiencing urbanization, which may not be well represented by the most recent national time-use survey from 1998 to 1999 [10
Global positioning system (GPS) technologies provide an alternative to self-reported data gathered via surveys or diaries to investigate daily mobility or time-activity patterns [11
]. The shortcomings of GPS (large amount of data, presence of missing data, battery and coverage issues) are outweighed by advantages such as low-cost, objectivity, low burden for participants, and high temporal precision. Automatic algorithms now detect locations, trips, and transportation modes from GPS tracks [12
]. GPS data have been used to investigate the impact of human mobility in infectious disease dynamics [16
], exposure to food environment [17
], levels of physical activity and sedentary behavior [18
], and to reduce misclassification in outdoor air pollution exposure assessment as compared to home-based estimates [2
The multidimensional nature of mobility makes it difficult to characterize thoroughly, while at the same time parsimoniously. Some previous studies have identified the main dimensions of mobility across multiple indicators based on a priori hypotheses or data-driven reduction techniques [20
], thereby facilitating the analysis of daily mobility in health studies. As part of the CHAI (Cardiovascular Health effects of Air pollution in Telangana, India) project, we collected daytime GPS tracks and time-activity diaries for one to six days per individual, during May 2015 to February 2016, for 60 adults living in a peri-urban area near Hyderabad, India. Our overall objective was to characterize objectively measured daily mobility of adults and explore how mobility differs between men and women. Specifically, we identified 1/the main dimensions of mobility across several indicators using principal component analysis, a data reduction technique, and 2/individual, external, and contextual predictors of these mobility dimensions.
The characteristics of the study population are presented in Table 2
. All participants reported Hindu religion. The mean (standard deviation, sd) age of the study population was 44 (14) years, with women being older than men. Women reported mostly unskilled manual occupation (71% vs. 39% for men), largely related to agriculture, and were less educated then men (79% and 26% illiteracy, respectively). Mean (sd) distance between residence and the nearest primary road was 4.4 (2.8) km. Other household characteristics along with village characteristics are presented in Supplementary Table S2
During daytime, participants spent an average (sd) of 11.5 (4.4) h at home, 3.2 (3.5) h in activity locations, and 1.6 (2.0) h in trips, but with major differences by sex. Women spent an average (sd) of 13.4 (3.7) h at home, compared to 9.4 (4.2) h for men (p
< 0.01), representing 83% and 57%, respectively, of their daytime (Figure 1
). Women spent six times more time at home than in other activity locations, while the corresponding ratio for men was only two. Mean daytime spent in trips was 2.8 h for men vs. only 0.6 h for women. Men made more trips than women (mean (sd) number of daily trips ≥5 min: 4.6 (3.5) vs. 1.5 (1.9), respectively) and men visited more activity locations than women (Table 3
). Women’s activity spaces were much smaller, more circular, and more home-centered than men’s (Supplementary Table S3
). Variability over time (i.e., within-participant) predominated for almost all mobility indicators. The most variable mobility indicators related to surface, perimeter, and centroid-to-home distance of the activity space with ICC ranging from 0.12 to 0.30, mainly because of male participants (Supplementary Table S4
In the study population, 72% and 14% of the points identified as “time spent at home” by the automated algorithm were consistently self-reported as “indoor at home” and “in playground or compound” in the hourly time-activity diary. This overall level of agreement was higher in women (76% and 16%, respectively) than in men (66% and 11%, respectively). In the hourly time-activity diary, men reported more “work” during daytime than women (27% vs. 15%, respectively). However, a similar proportion of points (~58%) identified as out of the residential place (namely, in activity places or in trips) by the automated algorithm were reported as “work” or “travel” by men and women.
The principal component analysis identified five and three dimensions of mobility, explaining 80% and 86% of the total variability in mobility for men and women, respectively (Table 4
). We labeled the dimensions according to the meaning of the indicators with the largest contribution. The size of the activity space dimension, with high contribution of the surface and perimeter of the activity space, was relevant for both men and women, but explained more variation in women (42%) than men (24%). Among men, the mobility in and around home dimension together with a dimension related to the circularity of the activity space explained 37% of the total variability in mobility. Among women, these two dimensions were grouped and explained a similar proportion of the total variability. Another common dimension to both sexes was that of mobility inside village, with a strong positive contribution of the proportion of activity locations visited inside village boundaries (Table 4
). These locations were most likely reached by walking by women (negative contribution of trip speed). The last dimension of men’s mobility explained almost 10% of the total variability, with the highest contribution of the median distance travelled from home. The results of the principal component analysis in men and women combined were similar to those found in men only and explained 84% of the total variability in mobility (Supplementary Table S5
). Using only data from the first session led to comparable mobility dimensions (Supplementary Table S6
shows the effects of individual, external, and contextual characteristics on the three main dimensions of mobility in men and women. Older men showed less mobility in and around home than younger men. The inverse association was observed among women, though not statistically significant. Being illiterate was associated with a smaller size of the activity space in men, and being illiterate was associated with less mobility in and around the home in women. Vehicle ownership (motorcycle or bicycle) led to a bigger size of the activity space in men, but a smaller size of the activity space in women. External factors (season and day type) showed effects on men’s mobility only (Figure 2
). Higher village-level urbanicity was associated with more mobility in and around home for both men and women. In men, higher village-level solid fuel use was also associated with more mobility in and around home. Increasing distance to the airport and to industry were associated with more mobility inside village in men and women. The count of non-residential places within residential buffers was, on the contrary, associated with less mobility inside village in women. Road length within residential buffers was associated with a bigger size of the activity space in women, but not men.
Our work addresses an important gap in the literature by describing daily mobility of adults living in a peri-urban area of India using objective GPS data. Our analysis provides three key findings. First, sex was a major determinant of mobility in this population for all indicators considered. Second, we successfully reduced the multidimensionality of mobility into a relatively small number of independent and meaningful dimensions. Third, we identified predictors (age, occupation, education level, vehicle ownership, day type, urbanicity, distance to airport, and distance to non-residential place) related to these mobility dimensions. Overall, our results provide new insights into adults’ mobility in an urbanizing area of India and have direct relevance for assessing exposure to various environmental hazards as well as health-related behaviors in future studies.
Daily mobility or time-activity patterns have been previously investigated, but comparisons with our findings are limited by different methodologies, research questions, study areas and populations. Nonetheless, our results are consistent with the literature regarding predictors of mobility, including: individual (sex, age, and socioeconomic position), contextual (urbanization level and land-use), and external (season and day type) factors.
Sex differences in mobility in the literature showed similar patterns as what we found here, but sex differences are considerably larger in our observations than in prior literature. In European and US cities, men spent only 1 to 1.6 h/day less time indoors at home than women [7
]. In the Delhi area, a time-budget survey reported men spent 5.6 h indoor at home (in the kitchen, drawing room, or bedroom), which was only 0.6 h less than women [33
]. In comparison, we observed a 4-h difference between men (9.4 h) and women (13.4 h) for time spent at home, which likely relates to the household burden carried by women and the associated occupation differences. Age differences in mobility found in the literature mostly related to the differences between childhood, working age, and elderly/retirement and generally showed less mobility with older age [7
]. In our population, age showed the opposite effect among men. That result could be explained by the high proportion of older men reporting an agriculture-related occupation. Having an agriculture-related occupation was indeed associated with less mobility in and around home, in particular less time spent at home, in men and women. This could relate to the increase time spent in trips which may reflect the multiple activities performed by agricultural workers. Consistent with other studies, we found that unemployed participants spent more time at the residence [7
]. However, the overall pattern across occupation levels was variable among men, which may be linked to their diverse activities and high variability in mobility. Motorized vehicle use has previously been related to the distance travelled from home and time spent in trips [20
]. In our population, household vehicle ownership showed an impact on all mobility dimensions, even among women who did not use such transportation mode (as showed by the maximum speed during trips: 6 km/h). Beyond access to transportation, household vehicle ownership likely is a good proxy for socioeconomic position in our population. Two-wheeled vehicle ownership within the household has been included in the Indian standard living index, among others indicators [40
]. Though the relevance of this index has been questioned in the face of the rapid urbanization, the specific vehicle ownership indicator may be still useful as a measure of socioeconomic position.
Urbanization, through the built environment and the availability of infrastructure, influences time spent travelling, distance travelled, and thus the size of the activity space in both high- [20
] and low-/middle-income countries [38
]. In a suburb of Chennai, India, the mean travel time was 50 min among women, consistent with the 36 min we found, and 1.4 h for men, which is less than the 2.7 h we found [43
]. This difference may be explained by the lower level of urbanization in our study area. The travel time in Chennai city center was estimated to be half that of the suburb [43
]. Similarly, we found that high village-level urbanicity related to less time spent in trips and more mobility in and around home. The limited spatial resolution of the village-level night-time light intensity may explain the observed lack of significant associations.
Typically for adults in high-income countries, time spent at home changed and activity locations differed from weekends to week days [35
]. In our population, this was only the case for men (more mobility in and around home and smaller and more circular activity space on weekends). Similarly, in a study in California, day type had stronger effect on men’s time-activity patterns than on women’s [35
]. Isaacs et al. found that day type, along with the season, were more important predictors than sex for time-activity patterns [44
]. In our population, the post-monsoon related to less time spent at home for women and bigger activity space for men, compared to other seasons. That observation likely relates to the increase in agricultural activities and the greater ease of travel compared to during the monsoon. In contrast, summer season relates to decreased agricultural activities and we observed smaller activity space and more mobility in and around home. According to men’s time-activity diary in our population, reporting indoor at home was lowest during summer season, while outdoor in playground or compound was highest (30% and 23% of daytime, respectively, vs. 45% and 7% in average during other seasons), supporting the idea that season might affect mobility through the balance between time spent indoor and outdoor.
We found a major effect of sex on daily mobility. A difference in mobility between men and women was partially expected because of the cultural context in India. The observed difference, however, was up to four times larger than any of the reported results in the literature. In our data, women stayed close to their residence (on average, they spent 13 h of their daytime within 50 m around home) and had much smaller and more circular activity spaces than men. Women’s mobility was characterized by walking speed trips, short trips, and a small number of activity locations. These findings may be explained by the high household burden carried by Indian women which restricts their mobility [45
]. A published analysis of the Indian time-use survey of 1998–1999 reported that women spent an estimated 33.5 h per week on household maintenance and care (vs. 2.9 h for men) [45
]. Considering a broader definition that included work inside and outside the home, Lahiri-dutt and Sil estimated household work to be 53.0 h per week for women in India [46
]. Women in our study location are predominantly engaged in informal, unpaid, or low-productivity work, which is usually home-based and intermittent, potentially explaining their limited mobility. The Indian time-use survey of 1998–1999 highlighted the predominance of female unpaid/informal workers in agricultural areas, which is consistent with our results [45
]. Though, independently of their occupation skill levels, women spent a large amount of time at home on average (>80% of daytime, Figure 1
). This large amount of time spent in the home vicinity suggests environmental exposures assessed at the residence may be effective estimates for women in this population, with the caveat that concentrations may differ between indoor and outdoor environments. In contrast, men’s mobility was more complex and variable than women’s. Specifically, men’s mobility variability over time was greater for activity locations indicators, time spent at home, and linear distance from home. The variability over time found in the present study was much larger than in high-income countries population [35
]. An important characteristic of the workforce in India is the multiple activities performed [45
], which may explain the variability over the course of the day and from one day to another in men. The high number of activity locations visited by men and the average of 3 h of trips also support this hypothesis. Another explanation might be the influence of external factors (day type and season) on men’s mobility. Day type and season thus may be important for environmental exposure assessment efforts in this or similar male populations.
Mobility is a multidimensional concept relevant for multiple disciplines and has been measured in different ways depending on the research question of interest. Trip-based definitions consider number of trips, travel time, and mode of transportation and are primarily used for urban planning, built environment, or health research focused on active transport [19
]. Activity space, the area containing all movements over a determined period of time, can be used to investigate the contextual physical or social effect of the neighborhood on health [47
]. Micro-environments, where an individual spends his/her time, are mainly used to understand environmental exposures [48
]. As previously proposed [20
], we identified the main dimensions of mobility across multiple existing indicators. In men and women, we identified three dimensions that were also identified in a similar analysis conducted among 2000 participants in Paris, France [20
]: the importance of the home vicinity, the size of the activity space, and the shape of the activity space. In the French study, these axes explained 64% of the total variability in mobility; in our population, they explained 61% (men) and 79% (women). Consistent with Perchoux et al., we also observed a dimension related to the volume of activities, although the indicators used were not directly comparable and the proportion of variance explained was lower in our study (<10% vs. 16%) [20
]. Indian women’s mobility showed high homogeneity and could be described along three main axes explaining 86% of the total variability. Men’s mobility showed similar features explaining almost 70% of the variability, though it was more complex with the shape of the activity space as a distinct dimension. The median distance travelled from the home dimension in men, needed to reach 80% of variability explained, lacked interpretability but might be related to the size of the activity space dimension. Overall, our results indicate that distinct and meaningful patterns of adult mobility can be identified in a peri-urban population of India. These patterns are likely to be informative in estimating environmental exposures or health behaviors in this or similar populations.
The strengths of our study include objectively measured mobility data using GPS devices, with repeated measurements over different seasons, involving a fairly minimal participant burden. Thanks to the longitudinal design and the homogeneity of the population, we were able to detect relevant associations despite the limited number of participants. The methods used throughout the present analysis (GPS-collected data, automated algorithm to derive activity locations, and data-reduction method) are innovative and generalizable. The principal components analysis identified indicators similar to those identified in an urban European population, suggesting that the relevant indicators to characterize mobility are also fairly generalizable across populations and levels of economic development. The specific findings regarding individual and contextual predictors of these mobility dimensions are likely to be generalizable across peri-urban and rural areas of South India (a population of approximately 250 million).
A number of limitations in the present study should be considered. The homogeneity of the population limited our ability to investigate modification of the relationship between predictors and mobility indicators by specific individual characteristics. We acknowledge that mobility was monitored for one session only for seven participants of the present study population. One day of data collection may not be sufficiently representative of the mobility habits of an individual. However, as our analysis primarily focused on the determinants of mobility in this population and not on individual variability over time, all data available were included. The automated algorithm we used could not distinguish between time spent indoors versus outdoors, a distinction potentially relevant for environmental exposure assessment. In high-income countries, time spent outdoors at home has been estimated to be only 3% in urban areas [7
]. Differences are expected for rural and suburban areas [36
], as well as for low- and middle-income countries. Using the time-activity diary in our population, we could estimate that 24% to 34% of the time spent at home was spent outdoors. The automated algorithm detected activity locations if time spent was >30 min [15
]. By excluding locations visited for short-durations, these were identified as trips. Thus, we likely underestimated the true number of activity locations visited. This aspect likely affected women’s mobility more than men’s, as women are expected to make short trips and stops related to household work and child care. We however believe that aspect did not impact our overall conclusions. Although we cannot exclude the possibility that participants changed their behaviors as a consequence of being monitored, field technicians have noticed that participants adapted quickly to the equipment, thus the influence of these behaviors on our conclusions might be limited. The prediction of transportation mode and identification of sedentary behavior were beyond the scope of the present study objectives. However, the CHAI project collected data from a collocated accelerometer, potentially making possible such analyses in the future.