Stress experiences in neighborhood and social environments (SENSE): a pilot study to integrate the quantified self with citizen science to improve the built environment and health

Stress experiences in neighborhood and social environments (SENSE): a pilot study to integrate... Background: Identifying elements of one’s environment—observable and unobservable—that contribute to chronic stress including the perception of comfort and discomfort associated with different settings, presents many methodo - logical and analytical challenges. However, it also presents an opportunity to engage the public in collecting and analyz- ing their own geospatial and biometric data to increase community member understanding of their local environments and activate potential environmental improvements. In this first-generation project, we developed a methodology to integrate geospatial technology with biometric sensing within a previously developed, evidence-based “citizen science” protocol, called “Our Voice.” Participants used a smartphone/tablet-based application, called the Discovery Tool (DT ), to collect photos and audio narratives about elements of the built environment that contributed to or detracted from their well-being. A wrist-worn sensor (Empatica E4) was used to collect time-stamped data, including 3-axis accelerometry, skin temperature, blood volume pressure, heart rate, heartbeat inter-beat interval, and electrodermal activity (EDA). Open-source R packages were employed to automatically organize, clean, geocode, and visualize the biometric data. Results: In total, 14 adults (8 women, 6 men) were successfully recruited to participate in the investigation. Partici- pants recorded 174 images and 124 audio files with the DT. Among captured images with a participant-determined positive or negative rating (n = 131), over half were positive (58.8%, n = 77). Within-participant positive/negative rating ratios were similar, with most participants rating 53.0% of their images as positive (SD 21.4%). Significant spatial clusters of positive and negative photos were identified using the Getis-Ord Gi* local statistic, and significant associa- tions between participant EDA and distance to DT photos, and street and land use characteristics were also observed with linear mixed models. Interactive data maps allowed participants to (1) reflect on data collected during the neigh- borhood walk, (2) see how EDA levels changed over the course of the walk in relation to objective neighborhood features (using basemap and DT app photos), and (3) compare their data to other participants along the same route. Conclusions: Participants identified a variety of social and environmental features that contributed to or detracted from their well-being. This initial investigation sets the stage for further research combining qualitative and quantita- tive data capture and interpretation to identify objective and perceived elements of the built environment influence our embodied experience in different settings. It provides a systematic process for simultaneously collecting multiple *Correspondence: bchris@stanford.edu Stanford Prevention Research Center, Department of Medicine, School of Medicine, Stanford University, 1070 Arastradero Road, Suite 100, Palo Alto, CA 94304, USA Full list of author information is available at the end of the article © The Author(s) 2018. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creat iveco mmons .org/licen ses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creat iveco mmons .org/ publi cdoma in/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Chrisinger and King Int J Health Geogr (2018) 17:17 Page 2 of 13 kinds of data, and lays a foundation for future statistical and spatial analyses in addition to more in-depth interpreta- tion of how these responses vary within and between individuals. Keywords: Citizen science, Quantified self, Chronic stress, Allostatic load, Electrodermal activity, Built environment, Sensors, Leaflet Background as cortisol or allostatic load allow for a consideration of Living and working in chronically stressful environments the cumulative effects of stress, they are not as well suited are thought to contribute to a wide range of adverse to understanding the relative influence of different stress - health outcomes. While the body’s stress response sys- ors, or how individual and environmental characteristics tems (e.g., “fight-or-flight”) may be helpful in adapting to intersect to yield different stress outcomes. acute environmental stimuli, major life events, or trauma, In response to this challenge, Roe and colleagues the continual triggering of these mechanisms may dimin- have spearheaded a new generation of interdiscipli- ish an individual’s biological capacity to respond to nary research that measures direct biological responses stressors [1]. With the impairment of this biological sys- to different types of built environments. Using a head- tem, chronic stress can increase an individual’s risk of worn device, the Emotiv EPOC, Roe et  al. [10] explored experiencing adverse health outcomes, including chronic changes in brain activity (measured via electroencepha- diseases such as obesity and type II diabetes [1–3]. logram, or EEG) as laboratory participants were exposed Emerging research is also beginning to help us conceptu- to images of urban and natural landscape scenes. These alize and document the causal pathways, including stress, findings informed later EEG studies with free-moving related to built environments that may explain the well- participants walking in different types of environments. documented burden of chronic disease in disadvantaged For instance, Aspinall et  al. [11] found EEG changes as communities [4–6]. participants moved into and out of urban green spaces Many studies exploring connections between stress and other environments. and health have used levels of blood or salivary cortisol, Complicating the identification of environmental con - a steroid hormone produced as part of the body’s stress tributors to chronic stress is the role of individual percep- response. Most cortisol research attempts to describe tion [12]. For instance, environmental health researchers cumulative stress effects, though additional attention to have found that individuals living in objectively noisy the different types of stressors—events, structural cir - neighborhoods do not, in general, exhibit the same ele- cumstances, or daily routines—is needed to understand vated stress responses to noise as those that do not. That possible pathways to poor health [7]. For example, prior is, individuals may become habituated to or learn to alter studies often have found limited or conflicting evidence their perceptions of such chronically stressful environ- about the nature of the relationship between cortisol and ments [13], although resident selection factors may also socioeconomic status [7]. McEwen and colleagues pro- be at work, at least to some extent (e.g., individuals who posed a broad framework for measuring chronic stress are particularly sensitive to high noise levels may move that included a set of ten biometric markers, includ- out of or avoid such neighborhoods) [14–19]. Based ing cortisol, a steroid hormone produced as part of the on such research, studies aimed at investigating physi- body’s stress response. Called “allostatic load” (AL), the ological responses to stress in differing environments measure was intended as a proxy measure for overall are challenged to integrate perception-based measures “wear and tear” on the body’s stress response systems within their built environment assessments. For example, [1]. Other researchers have found relationships between Aspinall et  al. [11] also asked participants to rate scenes AL and adverse mental and behavioral health outcomes, across several subjective criteria (e.g., pleasure/displeas- including cognition [2, 8]. ure, calm/excitement), adding a dimension of personal The concept of allostatic load has been applied to preference and underlying attitudes toward certain types chronic environmental stressors, especially those related of built environments. to workplace or neighborhood environments [8, 9]. One  relatively untested arena that may help us under- Theall et  al. [8] found that a significant amount of AL stand both individual characteristics and perceptions variance among adolescents in the National Health that matter for chronic stress is the quantified self. This and Nutrition Examination Survey was attributable to sensor-oriented movement encompasses a range of self- neighborhood-level factors, such as poverty, crime, and monitoring technologies, especially those that can be density of alcohol retailers, even after controlling for integrated into smartphones as mobile applications [20]. household-level characteristics. While biomarkers such Users typically collect and review biometric (e.g., weight, Chrisinger and King Int J Health Geogr (2018) 17:17 Page 3 of 13 heart rate) and/or behavioral (e.g., diet, sleep, exercise) inspection and interpretation, with some possible appli- data out of curiosity or an interest in self-improvement cations for future testing with additional spatial data. [20–22]. While quantified self applications may focus on The primary goals of this pilot study were to: (1) test individual or group-level changes in behavior through the feasibility of including biometric data collection via biofeedback or goal-setting, there is great potential to wrist-worn sensors as part of the objective and perceived develop relevant insights on population-level outcomes built environment data collection capabilities of the Dis- given sufficiently large datasets [23, 24]. For instance, a covery Tool mobile app being employed to obtain resi- recent smartphone-derived “big data” study found city- dent-collected information on local built environments; level correlations between objective walkability metrics (2) test the acceptability of different biometric data visu - and device-based walking outcomes measured from over alization styles; and (3) explore possible options for test- 700,000 smartphone users across 111 countries [24]. ing relationships between documented elements of the Aside from proprietary wearable datasets, some uni- built environment and biometric changes/outcomes. versity-sponsored projects aim to crowd-source place- based biometric data. One example is the Personal Methods Activity Location Monitoring System (PALMS), which Participant recruitment provides validated tools and methodologies for collecting To be eligible for the study, prospective participants had geo-located biometric data to track behavior across space to be healthy adult volunteers living in or around San and time, especially individual and group-level dynam- Francisco, California, and able to complete a relatively ics of physical activity [25–27]. Another suite of projects easy 20- to 25-min walk. Participants were recruited as employ “People as Sensors” methodologies that crowd- a convenience sample in San Francisco through the net- source a variety of objective and perceived data, includ- works of our San Francisco Bay Area community part- ing biometrics, in order to deliver relevant feedback to ner, an urban planning and design nonprofit called Place urban designers and planners [28–30]. For example, Zeile Lab, and through members of the research team. Walk and colleagues used a biosensor-oriented approach to appointments were scheduled for one of two days in July track how and why stress responses changed over space and September 2017. and time among a cohort of bicyclists in Cambridge, Massachusetts. Importantly, they found participants’ Biometric sensor mapped stress data corresponded to individual experi- A wrist-worn biometric sensor, the Empatica E4, was ences, as measured with video recordings [30]. selected for this study because it provides a commercially Our pilot study builds upon earlier biometric built available, easy-to-use means of continuously collecting environment assessments by integrating dimensions of time-stamped biometric data, including stress response the quantified self movement. First, we utilized a biom - and other possible measures that could be used in sub- etric measure of stress involving electrodermal activity sequent feature identification algorithms. Data collected (EDA), which has been shown to be an effective means of by the E4 include 3-axis accelerometry (which measures differentiating between different kinds of stress environ - gravitational force on three spatial dimensions, allow- ments and situations (e.g., driving in traffic vs. highway ing for a three-dimensional understanding of participant driving), and can be collected with relatively low impact movement), skin temperature, blood volume pressure, on participant experience compared to head- or chest- heart rate, heartbeat inter-beat interval, and electroder- worn devices [31]. Second, we embedded our biomet- mal activity (EDA). Once participants signed the study ric data collection within an existing successful “citizen consent form, which was approved by the Stanford Uni- science”-based community activation model, called “Our versity Institutional Review Board, they were asked to Voice”, which includes a mobile application, referred to put on and activate the sensor, which they wore during as the “Discovery Tool,” that allows community members a 10-min pre-walk period. The purpose of this approxi - to collect objective and perceived neighborhood data mately 10-min pre-walk data collection was two-fold: (1) [32, 33], providing a systematic and technology-assisted to allow for the sensor to make appropriate contact with enhancement to existing community-based qualitative the skin surface, and (2) to collect baseline electrodermal research methods, such as Photovoice [34, 35]. By creat- activity data for subsequent data analysis. ing a simple and reliable method of collecting geolocated stress data while participants use the Discovery Tool in Mobile built environment audit tool application the field, we aim to amplify the known strengths of this This investigation was intended to determine the initial type of citizen science model. Finally, we introduce open- feasibility and utility of adding biometric sensor data to source methods for visualizing and sharing perceived the built environment data collected with the DT app, and objective participant data with participants for their which is typically embedded within a broader citizen Chrisinger and King Int J Health Geogr (2018) 17:17 Page 4 of 13 science community engagement research method called and tablet (if borrowed) to the investigators. Five groups Our Voice. This method has been used successfully to participated over two separate days in August and Octo- study built and social environments in a variety of set- ber 2017, and group sizes ranged from one to four par- tings [32], and features the simple DT mobile application ticipants, depending on participant availability. to capture photos, audio narratives, and participant- Application stability issues related to the large quan- assigned valences for the specific built environment tity of photos and audio recordings taken by some par- elements being captured (“Is this [built environment ele- ticipants caused the DT app images from several walks ment] good or bad for the community?”). The DT app (n = 6) to be lost, though for two walks where audio files also captures the geocoordinates of the user’s walking were recovered, the research team was able re-create the route and a short demographic survey upon completion image in Google Street View by using the approximate of a walk. It additionally collects latitude and longitude location and subject being described in the audio file. coordinates every second while participants use it, and all For one of the participants without photo/audio data, the photos taken with the app has geo-coordinates embedded research team was also unable to recover biometric data. in their metadata. A web portal for viewing DT partici- Biometric data were successfully downloaded and pro- pant data allows the research team to download sum- cessed for the remaining participants (n = 13). mary data for each walk, including the walk route and locations of all photos and audio recordings. The full Our Data processing Voice process (not included in this pilot study) involves Each participant’s EDA data were normalized by sub- collection of geo-tagged photos and audio narratives with tracting the minimum value and dividing by the range the DT mobile app, followed by facilitated community from their baseline data values, consistent with prior meetings to identify shared themes and build community research using EDA data from the Empatica E4 biosensor consensus, in partnership with identified stakeholders, [31]. To assist with comparisons between participants, around how to address environmental issues negatively each participant’s normalized EDA data were also cen- impacting resident health and well-being and [32, 33, tered (subtracting the mean) and scaled (dividing by the 36–39]. standard deviation of the centered data). To help identify Depending on their preference, participants down- sudden changes, or “peaks,” in EDA, a proprietary algo- loaded the DT from the Apple App Store or Google Play rithm from the company was applied to help remove Store [40, 41], or used electronic tablets (Samsung Gal- erroneous readings, or “noise,” possibly caused by sudden axy Tab E Lite 7”) that were made available to them and motions or other non-EDA-related issues with the sensor. already contained the required DT app [42]. Participants Skin temperature, 3-axis accelerometry, and EDA data were verbally instructed on how to use the DT app, and files were provided as algorithm inputs; outputs included prompted to take photos and describe aspects of this time-stamped peaks in EDA with characteristics such as neighborhood environment that they felt influenced their peak amplitude and wavelengths. well-being or the functioning of these public spaces. Simple text processing was performed on participant audio narrative transcriptions using functions from the Neighborhood walks tm (“text mining”) package in R. To prepare the text for Based on our community partner’s interest in existing review, all letters were shifted to lowercase, very com- and future public space projects in a specific neighbor - mon words were removed, and a word frequency table hood of San Francisco, California, an approximately was generated [43, 44]. This table was further grouped by 20-min walking route (0.9  miles) was predetermined nouns and adjectives, and words with a frequency greater to take participants through a variety of different envi - than five were included in a visualization to compare the ronments, including a small public green space, back most prevalent terms across all participants. alleys, and busy commercial streets. Participants were instructed to document anything along the route that Data visualization they believed influenced their well-being or the function - Visualizations of each individual participant’s walking ing of these public spaces. A researcher accompanied route while using the DT app were generated with leaf- groups of up to four participants at a time to direct them let, an open-source JavaScript visualization library, which along the route and help troubleshoot any difficulties we deployed within the R software environment [45–47]. with the app or wearable sensor. Participants were also Markers indicating the location of photos/audio narra- encouraged not to talk to one another during the neigh- tives taken were added to these maps, and two-dimen- borhood walks. Following the neighborhood walk, partic- sional binned kernel density estimates were calculated to ipants completed a short demographic survey embedded visualize clusters of positive and negative photos (using within the DT app, and returned the biometric sensor the bkde2D function of the KernSmooth package) [48, Chrisinger and King Int J Health Geogr (2018) 17:17 Page 5 of 13 49]. The walking route was color-coded by participants’ highly-clustered cell, location within 10  m of a street relative EDA levels during their walk, with peaks illus- intersection, one/two-way traffic pattern and classifi - trated as additional markers of sizes according to their cation of street, land use of the nearest parcel, age of amplitude. These web-based visualizations were shared building on the nearest parcel, and the observation’s with participants via email, and their feedback was solic- time during the walk. A random intercept  was  specified ited with an open-ended web-based survey. Interactive to account for grouping of the study design: biometric data maps were generated for participant feedback, but observations within individuals (14 participants) within were not specifically analyzed as part of this study. These groups (5 walk groups). To illustrate possible within-sub- maps allowed participants to (1) reflect on data collected ject variations, simple linear regression models were fit during the neighborhood walk; (2) see how EDA levels for three participants for EDA outcomes and whether the changed over the course of the walk in relation to objec- observation was taken in positive or negative DT cluster. tive neighborhood features (e.g., basemap and DT app photos); and 3) compare their data to other participants along the same route. An example data map is shown in Results Fig.  1, and an interactive example  is provided as Addi- In total, 14 adults (8 women and 6 men) who lived in the tional file 1. San Francisco Bay Area were recruited to participate as a convenience sample. Participants recorded 181 images (mean 15.1, SD 8.4) and 146 audio files (mean 12.2, SD Spatial and statistical analyses 8.6) with the DT app, and 5416 geo-located biometric The database underlying the participant walk map visu - data observations were collected from 13 participants alizations described above was imported to a geographic (approximately 416 observations per participant). Fig- information system software, ArcMap 10.5.1 [50], where ure  2 illustrates the spatial distribution of photographs a variety of spatial data (listed below) from the City of taken with the DT app by the positive/negative valence San Francisco had been pre-assembled [51, 52]. Spatial assigned by participants for the built environment fea- joins were performed to assign each walking route GPS tures being captured. coordinate several fields from these local data, in addi - Among captured images that were tagged with a partic- tion to the biometric data used in the visualizations: ipant-coded valence (n = 131), just over half were positive distance to the nearest positive- and negative-rated DT (n = 77). Within-participant positive/negative valence photo, distance to the nearest street intersection (as pos- ratios were similar, with most participants rating 53.0% of sible points of interest in terms of high traffic/activity), their images as positive (SD 21.4%). The average distance parcel characteristics (e.g., current land use, age of build- to a positive DT photo during a walk was 11.9  m, while ings), and street characteristics (e.g., name, one/two-way the average distance to a negative photo was 14.7  m. In traffic pattern, classification as a highway, major, second - terms of the narratives participants used to explain their ary, or local street). photographs in the DT app, several words were repeated In addition to summary statistics of participants’ EDA frequently. The most common nouns used by partici - data by different spatial characteristics, two additional pants in their audio narratives included “street” (n = 35), analyses were performed. First, the significance of spa - “building” (n = 29), and “people” (n = 25), while “nice” tial clustering of positive and negative DT photos was (n = 22), “safe” (n = 11), and “good” (n = 10) were the assessed with the Optimized Hot Spot Analysis tool, most common adjectives. Figure  3 provides a visualiza- which calculates the Getis-Ord Gi* local statistic (Gi*), tion of all nouns and adjectives used by participants with a standardized measure of clustering for specified areal an overall frequency greater than five. units (here, set as 5-m grid cells along the walk path) Based on the Getis-Ord Gi* local statistic (estimated [53–55]. For this pilot, the top quintile of Gi* statistics at a 95% confidence level), two significant clusters of were selected as the most highly clustered cells; this pro- positive DT photos comprised approximately 4.3% of cedure was performed separately for positive and nega- the walk route (by distance), while negative photo clus- tive DT photos. A subsequent spatial join between these ters represented 2.7% of the route, also in two significant highly clustered cells and the participant walk data cre- clusters. Both of the significant positive clusters occurred ated a binary variable for observations inside or outside within the first half of the walk route, and both significant of a clustered cell. negative clusters occurred within the second half, though Second, a linear mixed model was fit on geo-located both negative and positive DT photos were taken on all biometric data observations using R (via lmer in the lme4 except one street during the course of the walk. Table  1 package) to identify associations between the main out- summarizes the frequency of walk observations by a vari- come, participant EDA, with contextual walk measures ety of environmental characteristics, including presence as fixed effects [56, 57]: location inside or outside of a Chrisinger and King Int J Health Geogr (2018) 17:17 Page 6 of 13 Fig. 1 Example of an interactive webpage built for participants to view and interpret their data. The red and blue markers show where this specific participant took photographs with the DT app. The participant’s path is color-coded by their EDA level, from dark purple to yellow (low to high). The complete html file and underlying R code has been uploaded as Additional files 1 and 2 inside of a positive or negative cluster, and Fig. 4 displays older buildings seeing higher observations, compared to summary statistics for electrodermal activity across these the most recent buildings. All models, 95% confidence characteristics. intervals, and coefficients are displayed in Table 2. Based on the linear mixed model, statistically signifi - Exploratory linear regression models for three indi- cant positive associations were found between partici- vidual participants also showed significant relationships pant EDA and positive photo clusters (B = 0.14, p < 0.001), for presence in a positive or negative DT photo cluster, and significant negative associations were found between though they were of varying magnitudes and directions participant EDA and negative photo clusters (B = − 0.17, (see Table  3). Additionally, these regressions indicated p < 0.001). This suggests that, on average, participants’ that positive and negative clusters had far better explana- EDA was higher in areas where many participants docu- tory value for some participants’ EDA than for others 2 2 mented more favorable features of the environment, and (e.g., R = 0.076 for Participant A3, vs. R = 0.263 for Par- lower in areas where they documented less favorable fea- ticipant B3). tures. Other significant associations between EDA and walk characteristics were also observed, including the Discussion street type (significantly lower for highway, major, and In this first-generation pilot investigation, we success - secondary streets compared to local streets, p < 0.001), fully assembled diverse technologies to collect and visu- and land uses compared to parcels designated as “open alize objective, perceived, and biometric data in an urban space” (significantly higher near mixed/residential neighborhood context. These data collection methods [p < 0.001] and residential [p = 0.033]; significantly lower provide researchers with a means of investigating both near office [p = 0.003] and vacant buildings [p < 0.001]). group and individual-level responses to different envi - The age of buildings also was associated with EDA, with ronmental conditions. Chrisinger and King Int J Health Geogr (2018) 17:17 Page 7 of 13 Density of All Positive and Negative Photos from the Discovery Tool App Discovery Tool Photos Positive Negative Unknown Significant Cluster Black & White Bla Open Str Op eet Map Satellite Sat Positive Cluster Po s 100 meters Negative Clusters Ne Significant Clusters Sig Wa Walk Path Discovery Dis Tool Data 100 meters Leaflet | Tiles © Esri — Source: Esri, i-cubed, USDA, USGS, AEX, GeoEye, Getmapping, Aerogrid, IGN, IGP, UPR-EGP, and the GIS User Com munity Fig. 2 Heat map of positive and negatively-rated photographs taken with the DT app. Colors indicate if clusters of photographs were rated as positive (blue shading) and negative (red shading) by participants using the DT app. Darker shading indicates a higher density, and yellow cells indicate clustering of positive/negative photos per the Getis-Ord Gi* local statistic (top quintile) In the case of the urban neighborhood walked by our document similar built environment features, this did participants, common perceptions of the built envi- not mean that their interpretations were identical. For ronment were observed, both in terms of the repeated example, as shown in Fig. 5, two participants on the same terms captured in DT narratives and the significant walk captured images of a particular intersection that clustering of positive and negative DT photos. At the did not allow pedestrians to cross on all sides. For one group level, linear mixed model testing confirmed that participant, this represented a barrier, while the other the average participant EDA levels observed inside a rated it as a positive feature. While this was not a com- positive cluster were significantly higher than those mon occurrence, it is possible that future studies with observed elsewhere on the walk. Conversely, EDA additional participants, better measures of participant observations inside negative DT photo clusters were demographics (e.g., age, gender, length of residence in/ significantly lower than those from outside them. Fur - around the city, etc.) or a focus on a narrower geographic ther exploration is needed to understand the dimen- area will find similar discrepancies between individual sions of this relationship, though these preliminary assessments of the same feature. Furthermore, as Table 3 statistical associations suggest that participants’ rat- illustrated, the strength of the relationship between EDA ings of the built environment were reflected in their and participant-rated DT photos may not be consistent stress responses. Additionally, significant correlations across all participants. Ultimately, this example under- between objective measures of the built environment, scores the importance of combining both objective and such as land use and street type, and EDA also high- perceived built environment information in assessments. light potentially influential relationships that could be more carefully tested with additional participant walks. Limitations These data also illustrated the value of multi-dimen - Another study limitation was the small number of resi- sional measurement at individual and group levels. While dents included in this initial feasibility study to develop at least some participants may have been motivated to a systematic process for collecting and analyzing diverse Chrisinger and King Int J Health Geogr (2018) 17:17 Page 8 of 13 Fig. 3 Visualization of nouns and adjectives with an overall frequency greater than five in all audio narratives types of data geo-located data. Though the participant thematic saturation when identifying positive and nega- sample is small, these individuals produced a relatively tive features of a particular built environment [32]. Addi- large database of qualitative data in terms of photo tionally, the fine-grained biometric data collection adds (n = 181) and audio narratives (n = 146). Similar studies thousands of additional data points in which these photos using the Discovery Tool application have found even and audio narratives can be contextualized. The methods small groups of participants (e.g., 8–10) are able to reach we have described are easily scalable to accommodate Chrisinger and King Int J Health Geogr (2018) 17:17 Page 9 of 13 Table 1 Summary of walk observations by environmental likely—and was observed by some participants—that characteristics individuals felt more inclined to take photographs when they observed others in their group doing so. One par- n obs. % Total obs. ticipant described this circumstance in an audio narra- DT clusters tive about a small church building along the route: “I find Inside positive 1122 21 that when the person I’m on a walk with takes a photo of Inside negative 1107 20 something, I want to take a photo of the same thing. But Street features it’s true, this blue building is pretty excellent” (see Addi- Intersection 416 8 tional file 3). One-way 3608 67 Another methodological question relates to the use Two-way 1804 33 of the Empatica E4 sensor. While we used a 10-min Local 2182 40 pre-walk period to obtain a reasonable baseline meas- Secondary 155 3 urement, it is possible that longer time periods or data Major 2619 48 collected under different circumstances (e.g., walk - Highway 456 8 ing vs. standing or sitting) may yield more complete Land use or interpretable measures for participants outside of Open space 315 6 laboratory environments. Other researchers using Culture/education 357 7 the Empatica sensor may find additional utility in the Mixed use 312 6 multi-dimensional data it creates to identify “signals” Mixed use/residential 1241 23 within the EDA data, versus “noise” possibly caused by Office 1716 32 a participant’s motions, perspiration due to exertion, or Industrial 329 6 other factors. The availability of such resources as the Residential 514 9 “EDA Explorer,” which integrates the sensor’s 3-axis Retail/entertainment 248 5 accelerometry, skin temperature, and EDA data to more Vacant 286 5 precisely estimate EDA changes, suggest that device Building age developers are considering the implications of collect- Post-1976 1604 30 ing electrodermal data in ambulatory settings, perhaps 1951–1975 196 4 a sign of future guidance on this topic [58, 59]. 1926–1950 1261 23 The neighborhood walk route was also the subject Pre-1925 1737 32 of methodological deliberations. While having partici- pants walk the same route allowed for a more direct comparison between participants’ DT app and biom- etric data, it also imposed a relatively arbitrary con- many more participants, should future researchers desire straint on what has often been a more free-form built to integrate them into crowd-sourcing initiatives, as in environment assessment in other Our Voice projects other quantified self projects [27]. [32]. Additionally, this pilot study allowed participants In terms of logistical challenges, we encountered issues to investigate a variety of urban spaces, though future with mobile app stability during some walks because of iterations may pursue a more in-depth exploration of a the amount of image/audio data being collected, which single, specific space, such as a park or plaza. Quanti - resulted in the loss of photo and audio data for some fied self researchers may also find utility in collecting participants. Our research team was able to retrieve or individual geo-tagged biometric stress data over several recreate some photos with Google Street View, and par- hours or days in future “n-of-1” studies or interventions ticipant experiences informed subsequent developer [60]. Importantly, the method described here is suffi - updates to the DT app (e.g., enabling reliable capture of ciently flexible to be tailored to the research questions larger quantities of images and audio narratives) along of new projects, but provides key capabilities for col- with an accompanying troubleshooting guide for new lecting and visualizing different kinds of objective and users. perceived participant data. Several methodological questions were also raised dur- Ultimately, the questions raised during the study ing the course of the pilot, and should be considered in may also prompt deeper qualitative analyses. As a par- future research. First, the potential effects of having par - ticipant eloquently summarized in one of her audio ticipants walk in groups, as undertaken in this study, ver- narratives: sus independently, are worth further consideration. It is Chrisinger and King Int J Health Geogr (2018) 17:17 Page 10 of 13 Table 2 Linear mixed model of participant EDA Participant EDA Characteristics in observations with group and participant-level random Different Walk Environments intercepts Mean Median Electrodermal activity (EDA) Positive DT Clusters B CI p Negative Intersection Fixed parts 1.Local_Street Time on walk 0.20 0.17 to 0.24 < .001 2.Secondary_Street Street 3.Major_Street Positive photo cluster 0.14 0.06 to 0.23 < .001 Features 4.Highway Negative photo cluster − 0.17 − 0.25 to − 0.09 < .001 One-Way Intersection 0.05 − 0.04 to 0.15 .284 Two-Way 2-way street (ref:1-way) 0.15 0.07 to 0.23 < .001 1.Open Space Street type (ref:Local) 2.Culture/Education 3.Mixed Use Secondary − 0.49 − 0.72 to − 0.27 < .001 4.Mixed Use/Resid. Major − 0.17 − 0.24 to − 0.11 < .001 Land 5.Office use Highway − 0.47 − 0.61 to − 0.34 < .001 6.Industrial Land use (ref: open space) 7.Residential Cultural − 0.11 − 0.31 to 0.08 .244 8.Retail/Entertain. Mixed 0.10 − 0.11 to 0.30 .365 9.Vacant 1.Post-1976 Mixed/residential 0.29 0.12 to 0.47 < .001 2.1951-1975 Building Office − 0.24 − 0.40 to − 0.08 .003 Year 3.1926-1950 Industrial 0.09 − 0.09 to 0.28 .329 4.Pre-1925 Residential 0.18 0.01 to 0.35 .033 -2 -1 0 1 2 Retail/entertainment 0.01 − 0.18 to 0.20 .924 Electrodermal Activity (EDA) Vacant − 0.38 − 0.55 to − 0.21 < .001 Fig. 4 Average and median EDA level observed by different walk Missing − 0.08 − 0.32 to 0.16 .498 environments (SD shown in brackets) Building age (ref: 1976–present) 1951–1975 0.15 − 0.03 to 0.33 .107 1926–1950 0.10 0.01 to 0.18 .030 There’s this interesting dichotomy that I don’t know Pre-1925 0.09 0.02 to 0.17 .011 how to express in a photograph which is the pleas- Unknown 0.32 0.16 to 0.48 < .001 ure of being in a complex urban environment bal- (Intercept) 0.01 − 0.19 to 0.20 .932 anced with a serenity and beauty. Both are pleas- Random parts ing, one is more intense which maybe might make σ 0.854 you… maybe my biometrics feel aggravated or N 14 disoriented in some ways but that is one of the rea- partid:(partgroup:time) N 5 sons we love cities so we should not optimize out a partgroup:time N 2 sense of complexity and chaos because that too is time Observations 5412 beautiful. 2 2 R /Ω .119/.119 Italic values are significant at p < 0.05 Table 3 Linear models of participant EDA observations showing within-subject correlations with positive/negative DT clusters Participant A3 Participant B3 Participant E3 B CI p B CI p B CI p (Intercept) 0.03 − 0.11 to 0.16 .680 − 0.23 − 0.35 to − 0.11 < .001 0.30 0.18 to 0.42 < .001 Positive cluster 0.35 0.10 to 0.60 .007 1.15 0.93 to 1.38 < .001 − 0.50 − 0.74 to − 0.26 < .001 Negative cluster − 0.47 − 0.71 to − 0.22 < .001 − 0.17 − 0.40 to 0.06 .147 − 1.00 − 1.22 to − 0.77 < .001 Observations 362 347 397 2 2 R/adj. R .076/.071 .263/.259 .164/.159 Italic values are significant at p < 0.05 Chrisinger and King Int J Health Geogr (2018) 17:17 Page 11 of 13 Fig. 5 Participant data maps. While these two participants also documented the same feature, they gave it different ratings and descriptions in terms of it being a positive or negative aspect of the built environment These pilot data provide a starting point for research - biometric data are also of interest to community mem- ers and citizen scientists to “triangulate” between the bers, and our open-source mapping technology (R code objective, perceived, and embodied experiences of provided as Additional file  2) allows for easier replica- built environment spaces in ways that could lead to tion in different settings and projects. It sets the stage new insights, including the beauty of “complexity and for additional research aimed at better understanding— chaos.” both qualitatively and quantitatively—how objective and perceived elements of the built environment influ - ence our “lived” experience in different settings, which may impact people’s stress as well as well-being and Conclusion quality of life. Identifying elements of one’s environment—both observable and unobservable—that contribute to allo- Additional files static load presents new opportunities to engage com- munity residents in collecting and analyzing their Additional file 1. Interactive Data Map. Geospatial visualization of partici- personal data to mobilize potential environmental pant data as an html file, suitable for viewing in a web browser. improvements. The current investigation provides a Additional file 2. R code. Sample R code for processing and visualizing systematic process of collecting these three types of participant Discovery Tool and biometric data. data, and lays a foundation for future spatial and statis Additional file 3. Example participant data maps. These two participants tical analyses in addition to more in-depth interpreta- were on the same walk and took photographs of the same building. One (at right) observed that they noticed this influence: “I find that when the tion of how these responses vary within and between person I’m on a walk with takes a photo of something, I want to take a participants. This type of multi-dimensional data col - photo of the same thing. But it’s true, this blue building is pretty excellent.” lection procedure could be integrated into future built environment or quantified self research projects where Chrisinger and King Int J Health Geogr (2018) 17:17 Page 12 of 13 Abbreviations Publisher’s Note AL: allostatic load; DT: Discovery Tool; EDA: electrodermal activity; EEG: electro- Springer Nature remains neutral with regard to jurisdictional claims in pub- encephalogram; SD: standard deviation. lished maps and institutional affiliations. Received: 29 January 2018 Accepted: 1 June 2018 Authors’ information Benjamin W. Chrisinger, Ph.D. is a Postdoctoral Research Fellow with the Stanford Prevention Research Center. With a background in urban planning and environmental sciences, Dr. Chrisinger uses his research to explore con- nections between the built environment and human health, especially health References disparities. Abby C. King, Ph.D. is a Professor of Health Research and Policy 1. McEwen BS, Stellar E. Stress and the individual: mechanisms leading to and of Medicine with the Stanford University School of Medicine. Dr. King is disease. Arch Intern Med. 1993;153:2093–101. a Recipient of the Outstanding Scientific Contributions in Health Psychology 2. Juster R-P, McEwen BS, Lupien SJ. Allostatic load biomarkers of chronic Award from the American Psychological Association. Her research focuses on stress and impact on health and cognition. Neurosci Biobehav Rev. the development, evaluation, and translation of public health interventions 2010;35:2–16. to reduce chronic disease. These research directions include expanding the 3. Lovallo WR. Stress and health: biological and psychological interactions. reach and generalizability of evidence-based interventions through use of Thousand Oaks: SAGE Publications; 2015. state-of-the-art communication technologies, community-based participa- 4. McCormack GR, Shiell A. In search of causality: a systematic review of the tory research perspectives to address health disparities among disadvantaged relationship between the built environment and physical activity among populations, and citizen science and policy-level approaches to health adults. Int J Behav Nutr Phys Act. 2011;8:125. promotion. 5. Drewnowski A, Aggarwal A, Tang W, Hurvitz PM, Scully J, Stewart O, et al. Obesity, diet quality, physical activity, and the built environment: the Author details need for behavioral pathways. BMC Public Health. 2016;16:1153. Stanford Prevention Research Center, Department of Medicine, School 6. Juarez PD, Matthews-Juarez P, Hood DB, Im W, Levine RS, Kilbourne BJ, of Medicine, Stanford University, 1070 Arastradero Road, Suite 100, Palo Alto, et al. The public health exposome: a population-based, exposure science CA 94304, USA. Department of Health Research and Policy, School of Medi- approach to health disparities research. Int J Environ Res Public Health. cine, Stanford University, Palo Alto, USA. 2014;11:12866–95. 7. Dowd JB, Simanek AM, Aiello AE. Socio-economic status, cortisol and allo- Authors’ contributions static load: a review of the literature. Int J Epidemiol. 2009;38:1297–309. BC and AK conceptualized and designed the study. BC executed the study 8. Theall KP, Drury SS, Shirtcliff EA. Cumulative neighborhood risk of and wrote the methodological code, and drafted the manuscript. AK provided psychosocial stress and allostatic load in adolescents. Am J Epidemiol. detailed feedback on the manuscript. Both authors read and approved the 2012;176:S164–74. final manuscript. 9. Keene DE, Geronimus AT. “Weathering” HOPE VI: the importance of evalu- ating the population health impact of public housing demolition and Acknowledgements displacement. J Urban Health Bull N Y Acad Med. 2011;88:417–35. The authors would like to acknowledge our community partner, Brooke Ray 10. Roe JJ, Aspinall PA, Mavros P, Coyne R. Engaging the brain: the impact of Rivera, Executive Director of Place Labs (formerly known as Build Public), for natural versus urban scenes using novel EEG methods in an experimental her collaborative spirit and assistance in recruiting and engaging participants setting. Environ Sci. 2013;1:93–104. in this study. We are also grateful to those community members who served 11. Aspinall P, Mavros P, Coyne R, Roe J. The urban brain: analysing outdoor as citizen scientists for this project. Finally, we would like to acknowledge physical activity with mobile EEG. Br J Sports Med 2013;49:272–6. several anonymous reviewers who provided constructive and insightful 12. Honold J, Beyer R, Lakes T, van der Meer E. Multiple environmental feedback. burdens and neighborhood-related health of city residents. J Environ Psychol. 2012;32:305–17. Competing interests 13. Hammer MS, Swinburn TK, Neitzel RL. Environmental noise pollution The authors declare that they have no competing interests. in the United States: developing an effective public health response. Environ Health Perspect. 2014;122:115–9. Availability of data and materials 14. Cohen S, Krantz DS, Evans GW, Stokols D. Community noise, behavior, The datasets used and/or analyzed during the current study are available from and health: the Los Angeles noise project. In: Baum A, Singer JE, editors. the corresponding author on reasonable request. Adv Environ Psychol [Internet]. Hillsdale, NJ: Erlbaum; 1982. p. 295–317. http://www.psy.cmu.edu/~scohe n/LAnoi sepro ject.pdf. Accessed 23 Jan Consent for publication As part of our participation consent form, all participants also gave consent for 15. Evans GW, Hygge S, Bullinger M. Chronic noise and psychological stress. the images, audio transcripts, and other non-identifiable data collected during Psychol Sci. 1995;6:333–8. this study to be published online (though only non-identifiable participant 16. Paneto GG, de Alvarez CE, Zannin PHT. Relationship between urban noise information was included in this manuscript). and the health of users of public spaces—a case study in Vitoria, ES, Brazil. J Build Constr Plan Res. 2017;5:45. Ethics approval and consent to participate 17. Stansfeld S, Haines M, Brown B. Noise and health in the urban environ- This study was performed in accordance with the Declaration of Helsinki and ment. Rev Environ Health. 2011;15:43–82. was approved on June 27, 2017 by the Stanford University Institutional Review 18. Paunović K, Jakovljević B, Belojević G. Predictors of noise annoyance in Board (Protocol #: 39736, IRB Assurance #: FWA00000935). noisy and quiet urban streets. Sci Total Environ. 2009;407:3707–11. 19. Kono S, Sone T. Residents’ response to environmental and neighborhood Funding noise. J Sound Vib. 1988;127:573–81. This work was supported by the Stanford Clinical and Translational Science 20. Swan M. Sensor mania! The internet of things, wearable computing, Award (CTSA) to Spectrum (UL1 TR001085). The CTSA program is led by the objective metrics, and the quantified self 2.0. J Sens Actuator Netw. National Center for Advancing Translational Sciences (NCATS) at the National 2012;1:217–53. Institutes of Health (NIH). The content is solely the responsibility of the authors 21. Whooley M, Ploderer B, Gray K. On the integration of self-tracking data and does not necessarily represent the official views of the NIH. Dr. Chrisinger amongst quantified self members. In: Proceedings of the 28th interna- was also supported by an NIH/NHLBI institutional postdoctoral training grant tional BCS human computer interaction conference HCI 2014-Sand Sea ( T32 HL007034). Sky-Holiday HCI. BCS; 2014. p. 151–60. Chrisinger and King Int J Health Geogr (2018) 17:17 Page 13 of 13 22. Shull PB, Jirattigalachote W, Hunt MA, Cutkosky MR, Delp SL. Quantified 40. Szeto I. Our voice discovery tool [Internet]. Irvin Szeto; 2017. https ://play. self and human movement: a review on the clinical impact of wearable googl e.com/store /apps/detai ls?id=edu.stanf ord.ourvo ice.disco veryt sensing and feedback for gait analysis and intervention. Gait Posture. ool&hl=en. Accessed 2 June 2018. 2014;40:11–9. 41. Discovery tool our voice on the app store [Internet]. App store. https :// 23. Barrett MA, Humblet O, Hiatt RA, Adler NE. Big data and disease itune s.apple .com/us/app/disco very-tool-our-voice /id117 19357 66?mt=8. prevention: from quantified self to quantified communities. Big Data. Accessed 7 Jan 2018. 2013;1:168–75. 42. Galaxy Tab E Lite 7.0ʺ 8 GB ( Wi-Fi) Tablets—SM-T113NDWAXAR | Sam- 24. Althoff T, Sosič R, Hicks JL, King AC, Delp SL, Leskovec J. Large-scale sung US [Internet]. Samsung Electron. Am. http://www.samsu ng.com/us/ physical activity data reveal worldwide activity inequality. Nature. mobil e/table ts/all-other -table ts/samsu ng-galax y-tab-e-lite-7-0-8gb-wi-fi- 2017;547:336.white -sm-t113n dwaxa r/. Accessed 7 Jan 2018. 25. Carlson JA, Jankowska MM, Meseck K, Godbole S, Natarajan L, Raab F, 43. Feinerer I, Hornik K, Artifex Software, Inc. tm: text mining package et al. Validity of PALMS GPS scoring of active and passive travel compared [Internet]. 2017. https ://cran.r-proje ct.org/web/packa ges/tm/index .html. to SenseCam. Med Sci Sports Exerc. 2015;47:662–7. Accessed 2 June 2018. 26. Ellis K, Godbole S, Kerr J, Lanckriet G. Multi-sensor physical activity rec- 44. Lang D, Chien G. wordcloud2: create word cloud by “htmlwidget” [Inter- ognition in free-living. In: Proceedings of ACM international conference net]. 2018. https ://cran.r-proje ct.org/web/packa ges/wordc loud2 /index ubiquitous computing; 2014. P. 431–40. .html. Accessed 2 June 2018. 27. UCSD-PALMS-Project—home [Internet]. https ://ucsd-palms -proje ct.wikis 45. Cheng J, Karambelkar B, Xie Y, Wickham H, Russell K, Johnson K, et al. paces .com/. Accessed 24 Apr 2018. Leaflet: create interactive web maps with the JavaScript “Leaflet” Library 28. Resch B. People as sensors and collective sensing-contextual observa- [Internet]. 2017. https ://cran.r-proje ct.org/web/packa ges/leafl et/index tions complementing geo-sensor network measurements. In: Krisp J, .html. Accessed 8 Jan 2018. editors. Progress in location-based services. Berlin: Springer; 2013. p. 46. Leaflet for R—introduction [Internet]. https ://rstud io.githu b.io/leafl et/. 391–406. Accessed 15 Mar 2018. 29. Zeile P, Resch B, Exner JP, Sagl G. Urban emotions: benefits and risks 47. Leaflet—an open-source JavaScript library for interactive maps [Internet]. in using human sensory assessment for the extraction of contextual http://leafl etjs.com/. Accessed 15 Mar 2018. emotion information in urban planning. In: Geertman S, Ferreira Jr. J, 48. Wand M, updates) BR (R port and. KernSmooth: functions for kernel Goodspeed R, Stillwell J, editors. Planning support systems and smart smoothing supporting Wand & Jones (1995) [Internet]. 2015. https :// cities. Cham: Springer; 2015. p. 209–25cran.r-proje ct.org/web/packa ges/KernS mooth /index .html. Accessed 2 30. Zeile P, Resch B, Loidl M, Petutschnig A, Dörrzapf L. Urban emotions and June 2018. cycling experience—enriching traffic planning for cyclists with human 49. Ruginski I. Visualizing interactive topographic maps using kernel density sensor data. GI_Forum 2016. 2016;1:204–16. in leaflet [Internet]. 2017. http://www.ianru ginsk i.com/visua lizin gtopo 31. Healey JA, Picard RW. Detecting stress during real-world driving tasks graph icmap s_tutor ial.html. Accessed 25 Jan 2018. using physiological sensors. IEEE Trans Intell Transp Syst. 2005;6:156–66. 50. ESRI. ArcMap 10.5.1. Redlands, CA: ESRI; 2017. 32. King AC, Winter SJ, Sheats JL, Rosas LG, Buman MP, Salvo D, et al. Leverag- 51. San Francisco Basemap Street Centerlines | DataSF | City and County of ing citizen science and information technology for population physical San Francisco [Internet]. https ://data.sfgov .org/Geogr aphic -Locat ions- activity promotion. Transl J Am Coll Sports Med. 2016;1:30–44.and-Bound aries /San-Franc isco-Basem ap-Stree t-Cente rline s/7hfy-8sz8/ 33. Buman MP, Winter SJ, Sheats JL, Hekler EB, Otten JJ, Grieco LA, et al. The about . Accessed 2 May 2018. Stanford Healthy Neighborhood Discovery Tool: a computerized tool to 52. Land Use | DataSF | City and County of San Francisco [Internet]. San Franc. assess active living environments. Am J Prev Med. 2013;44:e41–7. Data. https ://data.sfgov .org/Housi ng-and-Build ings/Land-Use/us3s-fp9q. 34. Wang CC, Cash JL, Powers LS. Who knows the streets as well as the Accessed 2 May 2018. homeless? Promoting personal and community action through photo- 53. Getis A, Ord JK. The analysis of spatial association by use of distance voice. Health Promot Pract. 2000;1:81–9. statistics. Geogr Anal. 1992;24:189–206. 35. Belon AP, Nieuwendyk LM, Vallianatos H, Nykiforuk CIJ. Perceived commu- 54. Ord JK, Getis A. Local spatial autocorrelation statistics: distributional issues nity environmental influences on eating behaviors: a photovoice analysis. and an application. Geogr Anal. 1995;27:286–306. Soc Sci Med. 1982;2016(171):18–29. 55. How Hot Spot Analysis (Getis-Ord Gi*) works—ArcGIS Pro | ArcGIS 36. Winter SJ, Rosas LG, Romero PP, Sheats JL, Buman MP, Baker C, et al. Using Desktop [Internet]. http://pro.arcgi s.com/en/pro-app/tool-refer ence/ citizen scientists to gather, analyze, and disseminate information about spati al-stati stics /h-how-hot-spot-analy sis-getis -ord-gi-spati al-stati .htm. neighborhood features that affect active living. J Immigr Minor Health. Accessed 27 Apr 2018. 2015;18(5):1126–38. 56. Bates D, Mächler M, Bolker B, Walker S. Fitting linear mixed-effects 37. Buman MP, Bertmann F, Hekler EB, Winter SJ, Sheats JL, King AC, et al. A models using lme4. ArXiv14065823 Stat [Internet]. 2014; http://arxiv .org/ qualitative study of shopper experiences at an urban farmers’ market abs/1406.5823. Accessed 2 June 2018. using the Stanford Healthy Neighborhood Discovery Tool. Public Health 57. Bates D, Mächler M, Bolker B, Walker S. Fitting linear mixed-effects models Nutr. 2015;18:994–1000. using lme4. J Stat Softw [Internet]; 067. https ://ideas .repec .org/a/jss/jstso 38. Sheats JL, Winter SJ, Romero PP, King AC. FEAST (Food Environment f/v067i 01.html. Accessed 1 Feb 2018. Assessment using the Stanford Tool): development of a mobile applica- 58. Taylor S, Jaques N, Chen W, Fedor S, Sano A, Picard R. Automatic identifi- tion to crowdsource resident interactions with the food environment. cation of artifacts in electrodermal activity data. In: EMBC. 2015. Ann Behav Med. 2014;47:(abstract). 59. EDA Explorer [Internet]. http://eda-explo rer.media .mit.edu/. Accessed 10 39. Chrisinger BW, Ramos A, Shaykis F, Martinez T, Banchoff AW, Winter SJ, Jan 2018. King AC. Leveraging citizen science for healthier food environments: a 60. Cushing CC, Walters RW, Hoffman L. Aggregated N-of-1 randomized con- pilot study to evaluate corner stores in Camden, New Jersey. Front Public trolled trials: modern data analytics applied to a clinically valid method of Health. 2018;6:89. intervention effectiveness. J Pediatr Psychol. 2014;39:138–50. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png International Journal of Health Geographics Springer Journals

Stress experiences in neighborhood and social environments (SENSE): a pilot study to integrate the quantified self with citizen science to improve the built environment and health

Free
13 pages

Loading next page...
 
/lp/springer_journal/stress-experiences-in-neighborhood-and-social-environments-sense-a-hvvBf6gnuD
Publisher
Springer Journals
Copyright
Copyright © 2018 by The Author(s)
Subject
Medicine & Public Health; Public Health; Geographical Information Systems/Cartography; Human Geography; Epidemiology; Remote Sensing/Photogrammetry; Health Promotion and Disease Prevention
eISSN
1476-072X
D.O.I.
10.1186/s12942-018-0140-1
Publisher site
See Article on Publisher Site

Abstract

Background: Identifying elements of one’s environment—observable and unobservable—that contribute to chronic stress including the perception of comfort and discomfort associated with different settings, presents many methodo - logical and analytical challenges. However, it also presents an opportunity to engage the public in collecting and analyz- ing their own geospatial and biometric data to increase community member understanding of their local environments and activate potential environmental improvements. In this first-generation project, we developed a methodology to integrate geospatial technology with biometric sensing within a previously developed, evidence-based “citizen science” protocol, called “Our Voice.” Participants used a smartphone/tablet-based application, called the Discovery Tool (DT ), to collect photos and audio narratives about elements of the built environment that contributed to or detracted from their well-being. A wrist-worn sensor (Empatica E4) was used to collect time-stamped data, including 3-axis accelerometry, skin temperature, blood volume pressure, heart rate, heartbeat inter-beat interval, and electrodermal activity (EDA). Open-source R packages were employed to automatically organize, clean, geocode, and visualize the biometric data. Results: In total, 14 adults (8 women, 6 men) were successfully recruited to participate in the investigation. Partici- pants recorded 174 images and 124 audio files with the DT. Among captured images with a participant-determined positive or negative rating (n = 131), over half were positive (58.8%, n = 77). Within-participant positive/negative rating ratios were similar, with most participants rating 53.0% of their images as positive (SD 21.4%). Significant spatial clusters of positive and negative photos were identified using the Getis-Ord Gi* local statistic, and significant associa- tions between participant EDA and distance to DT photos, and street and land use characteristics were also observed with linear mixed models. Interactive data maps allowed participants to (1) reflect on data collected during the neigh- borhood walk, (2) see how EDA levels changed over the course of the walk in relation to objective neighborhood features (using basemap and DT app photos), and (3) compare their data to other participants along the same route. Conclusions: Participants identified a variety of social and environmental features that contributed to or detracted from their well-being. This initial investigation sets the stage for further research combining qualitative and quantita- tive data capture and interpretation to identify objective and perceived elements of the built environment influence our embodied experience in different settings. It provides a systematic process for simultaneously collecting multiple *Correspondence: bchris@stanford.edu Stanford Prevention Research Center, Department of Medicine, School of Medicine, Stanford University, 1070 Arastradero Road, Suite 100, Palo Alto, CA 94304, USA Full list of author information is available at the end of the article © The Author(s) 2018. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creat iveco mmons .org/licen ses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creat iveco mmons .org/ publi cdoma in/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Chrisinger and King Int J Health Geogr (2018) 17:17 Page 2 of 13 kinds of data, and lays a foundation for future statistical and spatial analyses in addition to more in-depth interpreta- tion of how these responses vary within and between individuals. Keywords: Citizen science, Quantified self, Chronic stress, Allostatic load, Electrodermal activity, Built environment, Sensors, Leaflet Background as cortisol or allostatic load allow for a consideration of Living and working in chronically stressful environments the cumulative effects of stress, they are not as well suited are thought to contribute to a wide range of adverse to understanding the relative influence of different stress - health outcomes. While the body’s stress response sys- ors, or how individual and environmental characteristics tems (e.g., “fight-or-flight”) may be helpful in adapting to intersect to yield different stress outcomes. acute environmental stimuli, major life events, or trauma, In response to this challenge, Roe and colleagues the continual triggering of these mechanisms may dimin- have spearheaded a new generation of interdiscipli- ish an individual’s biological capacity to respond to nary research that measures direct biological responses stressors [1]. With the impairment of this biological sys- to different types of built environments. Using a head- tem, chronic stress can increase an individual’s risk of worn device, the Emotiv EPOC, Roe et  al. [10] explored experiencing adverse health outcomes, including chronic changes in brain activity (measured via electroencepha- diseases such as obesity and type II diabetes [1–3]. logram, or EEG) as laboratory participants were exposed Emerging research is also beginning to help us conceptu- to images of urban and natural landscape scenes. These alize and document the causal pathways, including stress, findings informed later EEG studies with free-moving related to built environments that may explain the well- participants walking in different types of environments. documented burden of chronic disease in disadvantaged For instance, Aspinall et  al. [11] found EEG changes as communities [4–6]. participants moved into and out of urban green spaces Many studies exploring connections between stress and other environments. and health have used levels of blood or salivary cortisol, Complicating the identification of environmental con - a steroid hormone produced as part of the body’s stress tributors to chronic stress is the role of individual percep- response. Most cortisol research attempts to describe tion [12]. For instance, environmental health researchers cumulative stress effects, though additional attention to have found that individuals living in objectively noisy the different types of stressors—events, structural cir - neighborhoods do not, in general, exhibit the same ele- cumstances, or daily routines—is needed to understand vated stress responses to noise as those that do not. That possible pathways to poor health [7]. For example, prior is, individuals may become habituated to or learn to alter studies often have found limited or conflicting evidence their perceptions of such chronically stressful environ- about the nature of the relationship between cortisol and ments [13], although resident selection factors may also socioeconomic status [7]. McEwen and colleagues pro- be at work, at least to some extent (e.g., individuals who posed a broad framework for measuring chronic stress are particularly sensitive to high noise levels may move that included a set of ten biometric markers, includ- out of or avoid such neighborhoods) [14–19]. Based ing cortisol, a steroid hormone produced as part of the on such research, studies aimed at investigating physi- body’s stress response. Called “allostatic load” (AL), the ological responses to stress in differing environments measure was intended as a proxy measure for overall are challenged to integrate perception-based measures “wear and tear” on the body’s stress response systems within their built environment assessments. For example, [1]. Other researchers have found relationships between Aspinall et  al. [11] also asked participants to rate scenes AL and adverse mental and behavioral health outcomes, across several subjective criteria (e.g., pleasure/displeas- including cognition [2, 8]. ure, calm/excitement), adding a dimension of personal The concept of allostatic load has been applied to preference and underlying attitudes toward certain types chronic environmental stressors, especially those related of built environments. to workplace or neighborhood environments [8, 9]. One  relatively untested arena that may help us under- Theall et  al. [8] found that a significant amount of AL stand both individual characteristics and perceptions variance among adolescents in the National Health that matter for chronic stress is the quantified self. This and Nutrition Examination Survey was attributable to sensor-oriented movement encompasses a range of self- neighborhood-level factors, such as poverty, crime, and monitoring technologies, especially those that can be density of alcohol retailers, even after controlling for integrated into smartphones as mobile applications [20]. household-level characteristics. While biomarkers such Users typically collect and review biometric (e.g., weight, Chrisinger and King Int J Health Geogr (2018) 17:17 Page 3 of 13 heart rate) and/or behavioral (e.g., diet, sleep, exercise) inspection and interpretation, with some possible appli- data out of curiosity or an interest in self-improvement cations for future testing with additional spatial data. [20–22]. While quantified self applications may focus on The primary goals of this pilot study were to: (1) test individual or group-level changes in behavior through the feasibility of including biometric data collection via biofeedback or goal-setting, there is great potential to wrist-worn sensors as part of the objective and perceived develop relevant insights on population-level outcomes built environment data collection capabilities of the Dis- given sufficiently large datasets [23, 24]. For instance, a covery Tool mobile app being employed to obtain resi- recent smartphone-derived “big data” study found city- dent-collected information on local built environments; level correlations between objective walkability metrics (2) test the acceptability of different biometric data visu - and device-based walking outcomes measured from over alization styles; and (3) explore possible options for test- 700,000 smartphone users across 111 countries [24]. ing relationships between documented elements of the Aside from proprietary wearable datasets, some uni- built environment and biometric changes/outcomes. versity-sponsored projects aim to crowd-source place- based biometric data. One example is the Personal Methods Activity Location Monitoring System (PALMS), which Participant recruitment provides validated tools and methodologies for collecting To be eligible for the study, prospective participants had geo-located biometric data to track behavior across space to be healthy adult volunteers living in or around San and time, especially individual and group-level dynam- Francisco, California, and able to complete a relatively ics of physical activity [25–27]. Another suite of projects easy 20- to 25-min walk. Participants were recruited as employ “People as Sensors” methodologies that crowd- a convenience sample in San Francisco through the net- source a variety of objective and perceived data, includ- works of our San Francisco Bay Area community part- ing biometrics, in order to deliver relevant feedback to ner, an urban planning and design nonprofit called Place urban designers and planners [28–30]. For example, Zeile Lab, and through members of the research team. Walk and colleagues used a biosensor-oriented approach to appointments were scheduled for one of two days in July track how and why stress responses changed over space and September 2017. and time among a cohort of bicyclists in Cambridge, Massachusetts. Importantly, they found participants’ Biometric sensor mapped stress data corresponded to individual experi- A wrist-worn biometric sensor, the Empatica E4, was ences, as measured with video recordings [30]. selected for this study because it provides a commercially Our pilot study builds upon earlier biometric built available, easy-to-use means of continuously collecting environment assessments by integrating dimensions of time-stamped biometric data, including stress response the quantified self movement. First, we utilized a biom - and other possible measures that could be used in sub- etric measure of stress involving electrodermal activity sequent feature identification algorithms. Data collected (EDA), which has been shown to be an effective means of by the E4 include 3-axis accelerometry (which measures differentiating between different kinds of stress environ - gravitational force on three spatial dimensions, allow- ments and situations (e.g., driving in traffic vs. highway ing for a three-dimensional understanding of participant driving), and can be collected with relatively low impact movement), skin temperature, blood volume pressure, on participant experience compared to head- or chest- heart rate, heartbeat inter-beat interval, and electroder- worn devices [31]. Second, we embedded our biomet- mal activity (EDA). Once participants signed the study ric data collection within an existing successful “citizen consent form, which was approved by the Stanford Uni- science”-based community activation model, called “Our versity Institutional Review Board, they were asked to Voice”, which includes a mobile application, referred to put on and activate the sensor, which they wore during as the “Discovery Tool,” that allows community members a 10-min pre-walk period. The purpose of this approxi - to collect objective and perceived neighborhood data mately 10-min pre-walk data collection was two-fold: (1) [32, 33], providing a systematic and technology-assisted to allow for the sensor to make appropriate contact with enhancement to existing community-based qualitative the skin surface, and (2) to collect baseline electrodermal research methods, such as Photovoice [34, 35]. By creat- activity data for subsequent data analysis. ing a simple and reliable method of collecting geolocated stress data while participants use the Discovery Tool in Mobile built environment audit tool application the field, we aim to amplify the known strengths of this This investigation was intended to determine the initial type of citizen science model. Finally, we introduce open- feasibility and utility of adding biometric sensor data to source methods for visualizing and sharing perceived the built environment data collected with the DT app, and objective participant data with participants for their which is typically embedded within a broader citizen Chrisinger and King Int J Health Geogr (2018) 17:17 Page 4 of 13 science community engagement research method called and tablet (if borrowed) to the investigators. Five groups Our Voice. This method has been used successfully to participated over two separate days in August and Octo- study built and social environments in a variety of set- ber 2017, and group sizes ranged from one to four par- tings [32], and features the simple DT mobile application ticipants, depending on participant availability. to capture photos, audio narratives, and participant- Application stability issues related to the large quan- assigned valences for the specific built environment tity of photos and audio recordings taken by some par- elements being captured (“Is this [built environment ele- ticipants caused the DT app images from several walks ment] good or bad for the community?”). The DT app (n = 6) to be lost, though for two walks where audio files also captures the geocoordinates of the user’s walking were recovered, the research team was able re-create the route and a short demographic survey upon completion image in Google Street View by using the approximate of a walk. It additionally collects latitude and longitude location and subject being described in the audio file. coordinates every second while participants use it, and all For one of the participants without photo/audio data, the photos taken with the app has geo-coordinates embedded research team was also unable to recover biometric data. in their metadata. A web portal for viewing DT partici- Biometric data were successfully downloaded and pro- pant data allows the research team to download sum- cessed for the remaining participants (n = 13). mary data for each walk, including the walk route and locations of all photos and audio recordings. The full Our Data processing Voice process (not included in this pilot study) involves Each participant’s EDA data were normalized by sub- collection of geo-tagged photos and audio narratives with tracting the minimum value and dividing by the range the DT mobile app, followed by facilitated community from their baseline data values, consistent with prior meetings to identify shared themes and build community research using EDA data from the Empatica E4 biosensor consensus, in partnership with identified stakeholders, [31]. To assist with comparisons between participants, around how to address environmental issues negatively each participant’s normalized EDA data were also cen- impacting resident health and well-being and [32, 33, tered (subtracting the mean) and scaled (dividing by the 36–39]. standard deviation of the centered data). To help identify Depending on their preference, participants down- sudden changes, or “peaks,” in EDA, a proprietary algo- loaded the DT from the Apple App Store or Google Play rithm from the company was applied to help remove Store [40, 41], or used electronic tablets (Samsung Gal- erroneous readings, or “noise,” possibly caused by sudden axy Tab E Lite 7”) that were made available to them and motions or other non-EDA-related issues with the sensor. already contained the required DT app [42]. Participants Skin temperature, 3-axis accelerometry, and EDA data were verbally instructed on how to use the DT app, and files were provided as algorithm inputs; outputs included prompted to take photos and describe aspects of this time-stamped peaks in EDA with characteristics such as neighborhood environment that they felt influenced their peak amplitude and wavelengths. well-being or the functioning of these public spaces. Simple text processing was performed on participant audio narrative transcriptions using functions from the Neighborhood walks tm (“text mining”) package in R. To prepare the text for Based on our community partner’s interest in existing review, all letters were shifted to lowercase, very com- and future public space projects in a specific neighbor - mon words were removed, and a word frequency table hood of San Francisco, California, an approximately was generated [43, 44]. This table was further grouped by 20-min walking route (0.9  miles) was predetermined nouns and adjectives, and words with a frequency greater to take participants through a variety of different envi - than five were included in a visualization to compare the ronments, including a small public green space, back most prevalent terms across all participants. alleys, and busy commercial streets. Participants were instructed to document anything along the route that Data visualization they believed influenced their well-being or the function - Visualizations of each individual participant’s walking ing of these public spaces. A researcher accompanied route while using the DT app were generated with leaf- groups of up to four participants at a time to direct them let, an open-source JavaScript visualization library, which along the route and help troubleshoot any difficulties we deployed within the R software environment [45–47]. with the app or wearable sensor. Participants were also Markers indicating the location of photos/audio narra- encouraged not to talk to one another during the neigh- tives taken were added to these maps, and two-dimen- borhood walks. Following the neighborhood walk, partic- sional binned kernel density estimates were calculated to ipants completed a short demographic survey embedded visualize clusters of positive and negative photos (using within the DT app, and returned the biometric sensor the bkde2D function of the KernSmooth package) [48, Chrisinger and King Int J Health Geogr (2018) 17:17 Page 5 of 13 49]. The walking route was color-coded by participants’ highly-clustered cell, location within 10  m of a street relative EDA levels during their walk, with peaks illus- intersection, one/two-way traffic pattern and classifi - trated as additional markers of sizes according to their cation of street, land use of the nearest parcel, age of amplitude. These web-based visualizations were shared building on the nearest parcel, and the observation’s with participants via email, and their feedback was solic- time during the walk. A random intercept  was  specified ited with an open-ended web-based survey. Interactive to account for grouping of the study design: biometric data maps were generated for participant feedback, but observations within individuals (14 participants) within were not specifically analyzed as part of this study. These groups (5 walk groups). To illustrate possible within-sub- maps allowed participants to (1) reflect on data collected ject variations, simple linear regression models were fit during the neighborhood walk; (2) see how EDA levels for three participants for EDA outcomes and whether the changed over the course of the walk in relation to objec- observation was taken in positive or negative DT cluster. tive neighborhood features (e.g., basemap and DT app photos); and 3) compare their data to other participants along the same route. An example data map is shown in Results Fig.  1, and an interactive example  is provided as Addi- In total, 14 adults (8 women and 6 men) who lived in the tional file 1. San Francisco Bay Area were recruited to participate as a convenience sample. Participants recorded 181 images (mean 15.1, SD 8.4) and 146 audio files (mean 12.2, SD Spatial and statistical analyses 8.6) with the DT app, and 5416 geo-located biometric The database underlying the participant walk map visu - data observations were collected from 13 participants alizations described above was imported to a geographic (approximately 416 observations per participant). Fig- information system software, ArcMap 10.5.1 [50], where ure  2 illustrates the spatial distribution of photographs a variety of spatial data (listed below) from the City of taken with the DT app by the positive/negative valence San Francisco had been pre-assembled [51, 52]. Spatial assigned by participants for the built environment fea- joins were performed to assign each walking route GPS tures being captured. coordinate several fields from these local data, in addi - Among captured images that were tagged with a partic- tion to the biometric data used in the visualizations: ipant-coded valence (n = 131), just over half were positive distance to the nearest positive- and negative-rated DT (n = 77). Within-participant positive/negative valence photo, distance to the nearest street intersection (as pos- ratios were similar, with most participants rating 53.0% of sible points of interest in terms of high traffic/activity), their images as positive (SD 21.4%). The average distance parcel characteristics (e.g., current land use, age of build- to a positive DT photo during a walk was 11.9  m, while ings), and street characteristics (e.g., name, one/two-way the average distance to a negative photo was 14.7  m. In traffic pattern, classification as a highway, major, second - terms of the narratives participants used to explain their ary, or local street). photographs in the DT app, several words were repeated In addition to summary statistics of participants’ EDA frequently. The most common nouns used by partici - data by different spatial characteristics, two additional pants in their audio narratives included “street” (n = 35), analyses were performed. First, the significance of spa - “building” (n = 29), and “people” (n = 25), while “nice” tial clustering of positive and negative DT photos was (n = 22), “safe” (n = 11), and “good” (n = 10) were the assessed with the Optimized Hot Spot Analysis tool, most common adjectives. Figure  3 provides a visualiza- which calculates the Getis-Ord Gi* local statistic (Gi*), tion of all nouns and adjectives used by participants with a standardized measure of clustering for specified areal an overall frequency greater than five. units (here, set as 5-m grid cells along the walk path) Based on the Getis-Ord Gi* local statistic (estimated [53–55]. For this pilot, the top quintile of Gi* statistics at a 95% confidence level), two significant clusters of were selected as the most highly clustered cells; this pro- positive DT photos comprised approximately 4.3% of cedure was performed separately for positive and nega- the walk route (by distance), while negative photo clus- tive DT photos. A subsequent spatial join between these ters represented 2.7% of the route, also in two significant highly clustered cells and the participant walk data cre- clusters. Both of the significant positive clusters occurred ated a binary variable for observations inside or outside within the first half of the walk route, and both significant of a clustered cell. negative clusters occurred within the second half, though Second, a linear mixed model was fit on geo-located both negative and positive DT photos were taken on all biometric data observations using R (via lmer in the lme4 except one street during the course of the walk. Table  1 package) to identify associations between the main out- summarizes the frequency of walk observations by a vari- come, participant EDA, with contextual walk measures ety of environmental characteristics, including presence as fixed effects [56, 57]: location inside or outside of a Chrisinger and King Int J Health Geogr (2018) 17:17 Page 6 of 13 Fig. 1 Example of an interactive webpage built for participants to view and interpret their data. The red and blue markers show where this specific participant took photographs with the DT app. The participant’s path is color-coded by their EDA level, from dark purple to yellow (low to high). The complete html file and underlying R code has been uploaded as Additional files 1 and 2 inside of a positive or negative cluster, and Fig. 4 displays older buildings seeing higher observations, compared to summary statistics for electrodermal activity across these the most recent buildings. All models, 95% confidence characteristics. intervals, and coefficients are displayed in Table 2. Based on the linear mixed model, statistically signifi - Exploratory linear regression models for three indi- cant positive associations were found between partici- vidual participants also showed significant relationships pant EDA and positive photo clusters (B = 0.14, p < 0.001), for presence in a positive or negative DT photo cluster, and significant negative associations were found between though they were of varying magnitudes and directions participant EDA and negative photo clusters (B = − 0.17, (see Table  3). Additionally, these regressions indicated p < 0.001). This suggests that, on average, participants’ that positive and negative clusters had far better explana- EDA was higher in areas where many participants docu- tory value for some participants’ EDA than for others 2 2 mented more favorable features of the environment, and (e.g., R = 0.076 for Participant A3, vs. R = 0.263 for Par- lower in areas where they documented less favorable fea- ticipant B3). tures. Other significant associations between EDA and walk characteristics were also observed, including the Discussion street type (significantly lower for highway, major, and In this first-generation pilot investigation, we success - secondary streets compared to local streets, p < 0.001), fully assembled diverse technologies to collect and visu- and land uses compared to parcels designated as “open alize objective, perceived, and biometric data in an urban space” (significantly higher near mixed/residential neighborhood context. These data collection methods [p < 0.001] and residential [p = 0.033]; significantly lower provide researchers with a means of investigating both near office [p = 0.003] and vacant buildings [p < 0.001]). group and individual-level responses to different envi - The age of buildings also was associated with EDA, with ronmental conditions. Chrisinger and King Int J Health Geogr (2018) 17:17 Page 7 of 13 Density of All Positive and Negative Photos from the Discovery Tool App Discovery Tool Photos Positive Negative Unknown Significant Cluster Black & White Bla Open Str Op eet Map Satellite Sat Positive Cluster Po s 100 meters Negative Clusters Ne Significant Clusters Sig Wa Walk Path Discovery Dis Tool Data 100 meters Leaflet | Tiles © Esri — Source: Esri, i-cubed, USDA, USGS, AEX, GeoEye, Getmapping, Aerogrid, IGN, IGP, UPR-EGP, and the GIS User Com munity Fig. 2 Heat map of positive and negatively-rated photographs taken with the DT app. Colors indicate if clusters of photographs were rated as positive (blue shading) and negative (red shading) by participants using the DT app. Darker shading indicates a higher density, and yellow cells indicate clustering of positive/negative photos per the Getis-Ord Gi* local statistic (top quintile) In the case of the urban neighborhood walked by our document similar built environment features, this did participants, common perceptions of the built envi- not mean that their interpretations were identical. For ronment were observed, both in terms of the repeated example, as shown in Fig. 5, two participants on the same terms captured in DT narratives and the significant walk captured images of a particular intersection that clustering of positive and negative DT photos. At the did not allow pedestrians to cross on all sides. For one group level, linear mixed model testing confirmed that participant, this represented a barrier, while the other the average participant EDA levels observed inside a rated it as a positive feature. While this was not a com- positive cluster were significantly higher than those mon occurrence, it is possible that future studies with observed elsewhere on the walk. Conversely, EDA additional participants, better measures of participant observations inside negative DT photo clusters were demographics (e.g., age, gender, length of residence in/ significantly lower than those from outside them. Fur - around the city, etc.) or a focus on a narrower geographic ther exploration is needed to understand the dimen- area will find similar discrepancies between individual sions of this relationship, though these preliminary assessments of the same feature. Furthermore, as Table 3 statistical associations suggest that participants’ rat- illustrated, the strength of the relationship between EDA ings of the built environment were reflected in their and participant-rated DT photos may not be consistent stress responses. Additionally, significant correlations across all participants. Ultimately, this example under- between objective measures of the built environment, scores the importance of combining both objective and such as land use and street type, and EDA also high- perceived built environment information in assessments. light potentially influential relationships that could be more carefully tested with additional participant walks. Limitations These data also illustrated the value of multi-dimen - Another study limitation was the small number of resi- sional measurement at individual and group levels. While dents included in this initial feasibility study to develop at least some participants may have been motivated to a systematic process for collecting and analyzing diverse Chrisinger and King Int J Health Geogr (2018) 17:17 Page 8 of 13 Fig. 3 Visualization of nouns and adjectives with an overall frequency greater than five in all audio narratives types of data geo-located data. Though the participant thematic saturation when identifying positive and nega- sample is small, these individuals produced a relatively tive features of a particular built environment [32]. Addi- large database of qualitative data in terms of photo tionally, the fine-grained biometric data collection adds (n = 181) and audio narratives (n = 146). Similar studies thousands of additional data points in which these photos using the Discovery Tool application have found even and audio narratives can be contextualized. The methods small groups of participants (e.g., 8–10) are able to reach we have described are easily scalable to accommodate Chrisinger and King Int J Health Geogr (2018) 17:17 Page 9 of 13 Table 1 Summary of walk observations by environmental likely—and was observed by some participants—that characteristics individuals felt more inclined to take photographs when they observed others in their group doing so. One par- n obs. % Total obs. ticipant described this circumstance in an audio narra- DT clusters tive about a small church building along the route: “I find Inside positive 1122 21 that when the person I’m on a walk with takes a photo of Inside negative 1107 20 something, I want to take a photo of the same thing. But Street features it’s true, this blue building is pretty excellent” (see Addi- Intersection 416 8 tional file 3). One-way 3608 67 Another methodological question relates to the use Two-way 1804 33 of the Empatica E4 sensor. While we used a 10-min Local 2182 40 pre-walk period to obtain a reasonable baseline meas- Secondary 155 3 urement, it is possible that longer time periods or data Major 2619 48 collected under different circumstances (e.g., walk - Highway 456 8 ing vs. standing or sitting) may yield more complete Land use or interpretable measures for participants outside of Open space 315 6 laboratory environments. Other researchers using Culture/education 357 7 the Empatica sensor may find additional utility in the Mixed use 312 6 multi-dimensional data it creates to identify “signals” Mixed use/residential 1241 23 within the EDA data, versus “noise” possibly caused by Office 1716 32 a participant’s motions, perspiration due to exertion, or Industrial 329 6 other factors. The availability of such resources as the Residential 514 9 “EDA Explorer,” which integrates the sensor’s 3-axis Retail/entertainment 248 5 accelerometry, skin temperature, and EDA data to more Vacant 286 5 precisely estimate EDA changes, suggest that device Building age developers are considering the implications of collect- Post-1976 1604 30 ing electrodermal data in ambulatory settings, perhaps 1951–1975 196 4 a sign of future guidance on this topic [58, 59]. 1926–1950 1261 23 The neighborhood walk route was also the subject Pre-1925 1737 32 of methodological deliberations. While having partici- pants walk the same route allowed for a more direct comparison between participants’ DT app and biom- etric data, it also imposed a relatively arbitrary con- many more participants, should future researchers desire straint on what has often been a more free-form built to integrate them into crowd-sourcing initiatives, as in environment assessment in other Our Voice projects other quantified self projects [27]. [32]. Additionally, this pilot study allowed participants In terms of logistical challenges, we encountered issues to investigate a variety of urban spaces, though future with mobile app stability during some walks because of iterations may pursue a more in-depth exploration of a the amount of image/audio data being collected, which single, specific space, such as a park or plaza. Quanti - resulted in the loss of photo and audio data for some fied self researchers may also find utility in collecting participants. Our research team was able to retrieve or individual geo-tagged biometric stress data over several recreate some photos with Google Street View, and par- hours or days in future “n-of-1” studies or interventions ticipant experiences informed subsequent developer [60]. Importantly, the method described here is suffi - updates to the DT app (e.g., enabling reliable capture of ciently flexible to be tailored to the research questions larger quantities of images and audio narratives) along of new projects, but provides key capabilities for col- with an accompanying troubleshooting guide for new lecting and visualizing different kinds of objective and users. perceived participant data. Several methodological questions were also raised dur- Ultimately, the questions raised during the study ing the course of the pilot, and should be considered in may also prompt deeper qualitative analyses. As a par- future research. First, the potential effects of having par - ticipant eloquently summarized in one of her audio ticipants walk in groups, as undertaken in this study, ver- narratives: sus independently, are worth further consideration. It is Chrisinger and King Int J Health Geogr (2018) 17:17 Page 10 of 13 Table 2 Linear mixed model of participant EDA Participant EDA Characteristics in observations with group and participant-level random Different Walk Environments intercepts Mean Median Electrodermal activity (EDA) Positive DT Clusters B CI p Negative Intersection Fixed parts 1.Local_Street Time on walk 0.20 0.17 to 0.24 < .001 2.Secondary_Street Street 3.Major_Street Positive photo cluster 0.14 0.06 to 0.23 < .001 Features 4.Highway Negative photo cluster − 0.17 − 0.25 to − 0.09 < .001 One-Way Intersection 0.05 − 0.04 to 0.15 .284 Two-Way 2-way street (ref:1-way) 0.15 0.07 to 0.23 < .001 1.Open Space Street type (ref:Local) 2.Culture/Education 3.Mixed Use Secondary − 0.49 − 0.72 to − 0.27 < .001 4.Mixed Use/Resid. Major − 0.17 − 0.24 to − 0.11 < .001 Land 5.Office use Highway − 0.47 − 0.61 to − 0.34 < .001 6.Industrial Land use (ref: open space) 7.Residential Cultural − 0.11 − 0.31 to 0.08 .244 8.Retail/Entertain. Mixed 0.10 − 0.11 to 0.30 .365 9.Vacant 1.Post-1976 Mixed/residential 0.29 0.12 to 0.47 < .001 2.1951-1975 Building Office − 0.24 − 0.40 to − 0.08 .003 Year 3.1926-1950 Industrial 0.09 − 0.09 to 0.28 .329 4.Pre-1925 Residential 0.18 0.01 to 0.35 .033 -2 -1 0 1 2 Retail/entertainment 0.01 − 0.18 to 0.20 .924 Electrodermal Activity (EDA) Vacant − 0.38 − 0.55 to − 0.21 < .001 Fig. 4 Average and median EDA level observed by different walk Missing − 0.08 − 0.32 to 0.16 .498 environments (SD shown in brackets) Building age (ref: 1976–present) 1951–1975 0.15 − 0.03 to 0.33 .107 1926–1950 0.10 0.01 to 0.18 .030 There’s this interesting dichotomy that I don’t know Pre-1925 0.09 0.02 to 0.17 .011 how to express in a photograph which is the pleas- Unknown 0.32 0.16 to 0.48 < .001 ure of being in a complex urban environment bal- (Intercept) 0.01 − 0.19 to 0.20 .932 anced with a serenity and beauty. Both are pleas- Random parts ing, one is more intense which maybe might make σ 0.854 you… maybe my biometrics feel aggravated or N 14 disoriented in some ways but that is one of the rea- partid:(partgroup:time) N 5 sons we love cities so we should not optimize out a partgroup:time N 2 sense of complexity and chaos because that too is time Observations 5412 beautiful. 2 2 R /Ω .119/.119 Italic values are significant at p < 0.05 Table 3 Linear models of participant EDA observations showing within-subject correlations with positive/negative DT clusters Participant A3 Participant B3 Participant E3 B CI p B CI p B CI p (Intercept) 0.03 − 0.11 to 0.16 .680 − 0.23 − 0.35 to − 0.11 < .001 0.30 0.18 to 0.42 < .001 Positive cluster 0.35 0.10 to 0.60 .007 1.15 0.93 to 1.38 < .001 − 0.50 − 0.74 to − 0.26 < .001 Negative cluster − 0.47 − 0.71 to − 0.22 < .001 − 0.17 − 0.40 to 0.06 .147 − 1.00 − 1.22 to − 0.77 < .001 Observations 362 347 397 2 2 R/adj. R .076/.071 .263/.259 .164/.159 Italic values are significant at p < 0.05 Chrisinger and King Int J Health Geogr (2018) 17:17 Page 11 of 13 Fig. 5 Participant data maps. While these two participants also documented the same feature, they gave it different ratings and descriptions in terms of it being a positive or negative aspect of the built environment These pilot data provide a starting point for research - biometric data are also of interest to community mem- ers and citizen scientists to “triangulate” between the bers, and our open-source mapping technology (R code objective, perceived, and embodied experiences of provided as Additional file  2) allows for easier replica- built environment spaces in ways that could lead to tion in different settings and projects. It sets the stage new insights, including the beauty of “complexity and for additional research aimed at better understanding— chaos.” both qualitatively and quantitatively—how objective and perceived elements of the built environment influ - ence our “lived” experience in different settings, which may impact people’s stress as well as well-being and Conclusion quality of life. Identifying elements of one’s environment—both observable and unobservable—that contribute to allo- Additional files static load presents new opportunities to engage com- munity residents in collecting and analyzing their Additional file 1. Interactive Data Map. Geospatial visualization of partici- personal data to mobilize potential environmental pant data as an html file, suitable for viewing in a web browser. improvements. The current investigation provides a Additional file 2. R code. Sample R code for processing and visualizing systematic process of collecting these three types of participant Discovery Tool and biometric data. data, and lays a foundation for future spatial and statis Additional file 3. Example participant data maps. These two participants tical analyses in addition to more in-depth interpreta- were on the same walk and took photographs of the same building. One (at right) observed that they noticed this influence: “I find that when the tion of how these responses vary within and between person I’m on a walk with takes a photo of something, I want to take a participants. This type of multi-dimensional data col - photo of the same thing. But it’s true, this blue building is pretty excellent.” lection procedure could be integrated into future built environment or quantified self research projects where Chrisinger and King Int J Health Geogr (2018) 17:17 Page 12 of 13 Abbreviations Publisher’s Note AL: allostatic load; DT: Discovery Tool; EDA: electrodermal activity; EEG: electro- Springer Nature remains neutral with regard to jurisdictional claims in pub- encephalogram; SD: standard deviation. lished maps and institutional affiliations. Received: 29 January 2018 Accepted: 1 June 2018 Authors’ information Benjamin W. Chrisinger, Ph.D. is a Postdoctoral Research Fellow with the Stanford Prevention Research Center. With a background in urban planning and environmental sciences, Dr. Chrisinger uses his research to explore con- nections between the built environment and human health, especially health References disparities. Abby C. King, Ph.D. is a Professor of Health Research and Policy 1. McEwen BS, Stellar E. Stress and the individual: mechanisms leading to and of Medicine with the Stanford University School of Medicine. Dr. King is disease. Arch Intern Med. 1993;153:2093–101. a Recipient of the Outstanding Scientific Contributions in Health Psychology 2. Juster R-P, McEwen BS, Lupien SJ. Allostatic load biomarkers of chronic Award from the American Psychological Association. Her research focuses on stress and impact on health and cognition. Neurosci Biobehav Rev. the development, evaluation, and translation of public health interventions 2010;35:2–16. to reduce chronic disease. These research directions include expanding the 3. Lovallo WR. Stress and health: biological and psychological interactions. reach and generalizability of evidence-based interventions through use of Thousand Oaks: SAGE Publications; 2015. state-of-the-art communication technologies, community-based participa- 4. McCormack GR, Shiell A. In search of causality: a systematic review of the tory research perspectives to address health disparities among disadvantaged relationship between the built environment and physical activity among populations, and citizen science and policy-level approaches to health adults. Int J Behav Nutr Phys Act. 2011;8:125. promotion. 5. Drewnowski A, Aggarwal A, Tang W, Hurvitz PM, Scully J, Stewart O, et al. Obesity, diet quality, physical activity, and the built environment: the Author details need for behavioral pathways. BMC Public Health. 2016;16:1153. Stanford Prevention Research Center, Department of Medicine, School 6. Juarez PD, Matthews-Juarez P, Hood DB, Im W, Levine RS, Kilbourne BJ, of Medicine, Stanford University, 1070 Arastradero Road, Suite 100, Palo Alto, et al. The public health exposome: a population-based, exposure science CA 94304, USA. Department of Health Research and Policy, School of Medi- approach to health disparities research. Int J Environ Res Public Health. cine, Stanford University, Palo Alto, USA. 2014;11:12866–95. 7. Dowd JB, Simanek AM, Aiello AE. Socio-economic status, cortisol and allo- Authors’ contributions static load: a review of the literature. Int J Epidemiol. 2009;38:1297–309. BC and AK conceptualized and designed the study. BC executed the study 8. Theall KP, Drury SS, Shirtcliff EA. Cumulative neighborhood risk of and wrote the methodological code, and drafted the manuscript. AK provided psychosocial stress and allostatic load in adolescents. Am J Epidemiol. detailed feedback on the manuscript. Both authors read and approved the 2012;176:S164–74. final manuscript. 9. Keene DE, Geronimus AT. “Weathering” HOPE VI: the importance of evalu- ating the population health impact of public housing demolition and Acknowledgements displacement. J Urban Health Bull N Y Acad Med. 2011;88:417–35. The authors would like to acknowledge our community partner, Brooke Ray 10. Roe JJ, Aspinall PA, Mavros P, Coyne R. Engaging the brain: the impact of Rivera, Executive Director of Place Labs (formerly known as Build Public), for natural versus urban scenes using novel EEG methods in an experimental her collaborative spirit and assistance in recruiting and engaging participants setting. Environ Sci. 2013;1:93–104. in this study. We are also grateful to those community members who served 11. Aspinall P, Mavros P, Coyne R, Roe J. The urban brain: analysing outdoor as citizen scientists for this project. Finally, we would like to acknowledge physical activity with mobile EEG. Br J Sports Med 2013;49:272–6. several anonymous reviewers who provided constructive and insightful 12. Honold J, Beyer R, Lakes T, van der Meer E. Multiple environmental feedback. burdens and neighborhood-related health of city residents. J Environ Psychol. 2012;32:305–17. Competing interests 13. Hammer MS, Swinburn TK, Neitzel RL. Environmental noise pollution The authors declare that they have no competing interests. in the United States: developing an effective public health response. Environ Health Perspect. 2014;122:115–9. Availability of data and materials 14. Cohen S, Krantz DS, Evans GW, Stokols D. Community noise, behavior, The datasets used and/or analyzed during the current study are available from and health: the Los Angeles noise project. In: Baum A, Singer JE, editors. the corresponding author on reasonable request. Adv Environ Psychol [Internet]. Hillsdale, NJ: Erlbaum; 1982. p. 295–317. http://www.psy.cmu.edu/~scohe n/LAnoi sepro ject.pdf. Accessed 23 Jan Consent for publication As part of our participation consent form, all participants also gave consent for 15. Evans GW, Hygge S, Bullinger M. Chronic noise and psychological stress. the images, audio transcripts, and other non-identifiable data collected during Psychol Sci. 1995;6:333–8. this study to be published online (though only non-identifiable participant 16. Paneto GG, de Alvarez CE, Zannin PHT. Relationship between urban noise information was included in this manuscript). and the health of users of public spaces—a case study in Vitoria, ES, Brazil. J Build Constr Plan Res. 2017;5:45. Ethics approval and consent to participate 17. Stansfeld S, Haines M, Brown B. Noise and health in the urban environ- This study was performed in accordance with the Declaration of Helsinki and ment. Rev Environ Health. 2011;15:43–82. was approved on June 27, 2017 by the Stanford University Institutional Review 18. Paunović K, Jakovljević B, Belojević G. Predictors of noise annoyance in Board (Protocol #: 39736, IRB Assurance #: FWA00000935). noisy and quiet urban streets. Sci Total Environ. 2009;407:3707–11. 19. Kono S, Sone T. Residents’ response to environmental and neighborhood Funding noise. J Sound Vib. 1988;127:573–81. This work was supported by the Stanford Clinical and Translational Science 20. Swan M. Sensor mania! The internet of things, wearable computing, Award (CTSA) to Spectrum (UL1 TR001085). The CTSA program is led by the objective metrics, and the quantified self 2.0. J Sens Actuator Netw. National Center for Advancing Translational Sciences (NCATS) at the National 2012;1:217–53. Institutes of Health (NIH). The content is solely the responsibility of the authors 21. Whooley M, Ploderer B, Gray K. On the integration of self-tracking data and does not necessarily represent the official views of the NIH. Dr. Chrisinger amongst quantified self members. In: Proceedings of the 28th interna- was also supported by an NIH/NHLBI institutional postdoctoral training grant tional BCS human computer interaction conference HCI 2014-Sand Sea ( T32 HL007034). Sky-Holiday HCI. BCS; 2014. p. 151–60. Chrisinger and King Int J Health Geogr (2018) 17:17 Page 13 of 13 22. Shull PB, Jirattigalachote W, Hunt MA, Cutkosky MR, Delp SL. Quantified 40. Szeto I. Our voice discovery tool [Internet]. Irvin Szeto; 2017. https ://play. self and human movement: a review on the clinical impact of wearable googl e.com/store /apps/detai ls?id=edu.stanf ord.ourvo ice.disco veryt sensing and feedback for gait analysis and intervention. Gait Posture. ool&hl=en. Accessed 2 June 2018. 2014;40:11–9. 41. Discovery tool our voice on the app store [Internet]. App store. https :// 23. Barrett MA, Humblet O, Hiatt RA, Adler NE. Big data and disease itune s.apple .com/us/app/disco very-tool-our-voice /id117 19357 66?mt=8. prevention: from quantified self to quantified communities. Big Data. Accessed 7 Jan 2018. 2013;1:168–75. 42. Galaxy Tab E Lite 7.0ʺ 8 GB ( Wi-Fi) Tablets—SM-T113NDWAXAR | Sam- 24. Althoff T, Sosič R, Hicks JL, King AC, Delp SL, Leskovec J. Large-scale sung US [Internet]. Samsung Electron. Am. http://www.samsu ng.com/us/ physical activity data reveal worldwide activity inequality. Nature. mobil e/table ts/all-other -table ts/samsu ng-galax y-tab-e-lite-7-0-8gb-wi-fi- 2017;547:336.white -sm-t113n dwaxa r/. Accessed 7 Jan 2018. 25. Carlson JA, Jankowska MM, Meseck K, Godbole S, Natarajan L, Raab F, 43. Feinerer I, Hornik K, Artifex Software, Inc. tm: text mining package et al. Validity of PALMS GPS scoring of active and passive travel compared [Internet]. 2017. https ://cran.r-proje ct.org/web/packa ges/tm/index .html. to SenseCam. Med Sci Sports Exerc. 2015;47:662–7. Accessed 2 June 2018. 26. Ellis K, Godbole S, Kerr J, Lanckriet G. Multi-sensor physical activity rec- 44. Lang D, Chien G. wordcloud2: create word cloud by “htmlwidget” [Inter- ognition in free-living. In: Proceedings of ACM international conference net]. 2018. https ://cran.r-proje ct.org/web/packa ges/wordc loud2 /index ubiquitous computing; 2014. P. 431–40. .html. Accessed 2 June 2018. 27. UCSD-PALMS-Project—home [Internet]. https ://ucsd-palms -proje ct.wikis 45. Cheng J, Karambelkar B, Xie Y, Wickham H, Russell K, Johnson K, et al. paces .com/. Accessed 24 Apr 2018. Leaflet: create interactive web maps with the JavaScript “Leaflet” Library 28. Resch B. People as sensors and collective sensing-contextual observa- [Internet]. 2017. https ://cran.r-proje ct.org/web/packa ges/leafl et/index tions complementing geo-sensor network measurements. In: Krisp J, .html. Accessed 8 Jan 2018. editors. Progress in location-based services. Berlin: Springer; 2013. p. 46. Leaflet for R—introduction [Internet]. https ://rstud io.githu b.io/leafl et/. 391–406. Accessed 15 Mar 2018. 29. Zeile P, Resch B, Exner JP, Sagl G. Urban emotions: benefits and risks 47. Leaflet—an open-source JavaScript library for interactive maps [Internet]. in using human sensory assessment for the extraction of contextual http://leafl etjs.com/. Accessed 15 Mar 2018. emotion information in urban planning. In: Geertman S, Ferreira Jr. J, 48. Wand M, updates) BR (R port and. KernSmooth: functions for kernel Goodspeed R, Stillwell J, editors. Planning support systems and smart smoothing supporting Wand & Jones (1995) [Internet]. 2015. https :// cities. Cham: Springer; 2015. p. 209–25cran.r-proje ct.org/web/packa ges/KernS mooth /index .html. Accessed 2 30. Zeile P, Resch B, Loidl M, Petutschnig A, Dörrzapf L. Urban emotions and June 2018. cycling experience—enriching traffic planning for cyclists with human 49. Ruginski I. Visualizing interactive topographic maps using kernel density sensor data. GI_Forum 2016. 2016;1:204–16. in leaflet [Internet]. 2017. http://www.ianru ginsk i.com/visua lizin gtopo 31. Healey JA, Picard RW. Detecting stress during real-world driving tasks graph icmap s_tutor ial.html. Accessed 25 Jan 2018. using physiological sensors. IEEE Trans Intell Transp Syst. 2005;6:156–66. 50. ESRI. ArcMap 10.5.1. Redlands, CA: ESRI; 2017. 32. King AC, Winter SJ, Sheats JL, Rosas LG, Buman MP, Salvo D, et al. Leverag- 51. San Francisco Basemap Street Centerlines | DataSF | City and County of ing citizen science and information technology for population physical San Francisco [Internet]. https ://data.sfgov .org/Geogr aphic -Locat ions- activity promotion. Transl J Am Coll Sports Med. 2016;1:30–44.and-Bound aries /San-Franc isco-Basem ap-Stree t-Cente rline s/7hfy-8sz8/ 33. Buman MP, Winter SJ, Sheats JL, Hekler EB, Otten JJ, Grieco LA, et al. The about . Accessed 2 May 2018. Stanford Healthy Neighborhood Discovery Tool: a computerized tool to 52. Land Use | DataSF | City and County of San Francisco [Internet]. San Franc. assess active living environments. Am J Prev Med. 2013;44:e41–7. Data. https ://data.sfgov .org/Housi ng-and-Build ings/Land-Use/us3s-fp9q. 34. Wang CC, Cash JL, Powers LS. Who knows the streets as well as the Accessed 2 May 2018. homeless? Promoting personal and community action through photo- 53. Getis A, Ord JK. The analysis of spatial association by use of distance voice. Health Promot Pract. 2000;1:81–9. statistics. Geogr Anal. 1992;24:189–206. 35. Belon AP, Nieuwendyk LM, Vallianatos H, Nykiforuk CIJ. Perceived commu- 54. Ord JK, Getis A. Local spatial autocorrelation statistics: distributional issues nity environmental influences on eating behaviors: a photovoice analysis. and an application. Geogr Anal. 1995;27:286–306. Soc Sci Med. 1982;2016(171):18–29. 55. How Hot Spot Analysis (Getis-Ord Gi*) works—ArcGIS Pro | ArcGIS 36. Winter SJ, Rosas LG, Romero PP, Sheats JL, Buman MP, Baker C, et al. Using Desktop [Internet]. http://pro.arcgi s.com/en/pro-app/tool-refer ence/ citizen scientists to gather, analyze, and disseminate information about spati al-stati stics /h-how-hot-spot-analy sis-getis -ord-gi-spati al-stati .htm. neighborhood features that affect active living. J Immigr Minor Health. Accessed 27 Apr 2018. 2015;18(5):1126–38. 56. Bates D, Mächler M, Bolker B, Walker S. Fitting linear mixed-effects 37. Buman MP, Bertmann F, Hekler EB, Winter SJ, Sheats JL, King AC, et al. A models using lme4. ArXiv14065823 Stat [Internet]. 2014; http://arxiv .org/ qualitative study of shopper experiences at an urban farmers’ market abs/1406.5823. Accessed 2 June 2018. using the Stanford Healthy Neighborhood Discovery Tool. Public Health 57. Bates D, Mächler M, Bolker B, Walker S. Fitting linear mixed-effects models Nutr. 2015;18:994–1000. using lme4. J Stat Softw [Internet]; 067. https ://ideas .repec .org/a/jss/jstso 38. Sheats JL, Winter SJ, Romero PP, King AC. FEAST (Food Environment f/v067i 01.html. Accessed 1 Feb 2018. Assessment using the Stanford Tool): development of a mobile applica- 58. Taylor S, Jaques N, Chen W, Fedor S, Sano A, Picard R. Automatic identifi- tion to crowdsource resident interactions with the food environment. cation of artifacts in electrodermal activity data. In: EMBC. 2015. Ann Behav Med. 2014;47:(abstract). 59. EDA Explorer [Internet]. http://eda-explo rer.media .mit.edu/. Accessed 10 39. Chrisinger BW, Ramos A, Shaykis F, Martinez T, Banchoff AW, Winter SJ, Jan 2018. King AC. Leveraging citizen science for healthier food environments: a 60. Cushing CC, Walters RW, Hoffman L. Aggregated N-of-1 randomized con- pilot study to evaluate corner stores in Camden, New Jersey. Front Public trolled trials: modern data analytics applied to a clinically valid method of Health. 2018;6:89. intervention effectiveness. J Pediatr Psychol. 2014;39:138–50.

Journal

International Journal of Health GeographicsSpringer Journals

Published: Jun 5, 2018

References

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off