Data Engagement Reconsidered: A Study of Automatic Stress Tracking Technology in Use
DData Engagement Reconsidered: A Study of Automatic StressTracking Technology in Use
Xianghua Ding [email protected] Key Laboratory of DataScience, School of Computer ScienceFudan UniversityShanghai, China
Shuhan Wei
Xinning Gui [email protected] of Information Sciences andTechnologyPennsylvania State UniversityUSA
Ning Gu [email protected] of Computer ScienceFudan UniversityShanghai, China
Peng Zhang [email protected] of Computer ScienceFudan UniversityShanghai, China
ABSTRACT
In today’s fast-paced world, stress has become a growing healthconcern. While more automatic stress tracking technologies have re-cently become available on wearable or mobile devices, there is stilla limited understanding of how they are actually used in everydaylife. This paper presents an empirical study of automatic stress-tracking technologies in use in China, based on semi-structuredinterviews with 17 users. The study highlights three challengesof stress-tracking data engagement that prevent effective tech-nology usage: the lack of immediate awareness, the lack of pre-required knowledge, and the lack of corresponding communal sup-port. Drawing on the stress-tracking practices uncovered in thestudy, we bring these issues to the fore, and unpack assumptionsembedded in related works on self-tracking and how data engage-ment is approached. We end by calling for a reconsideration of dataengagement as part of self-tracking practices with technologiesrather than simply looking at the user interface.
CCS CONCEPTS • Human-centered computing → Empirical studies in HCI . KEYWORDS self-tracking, stress-tracking, knowing
ACM Reference Format:
Xianghua Ding, Shuhan Wei, Xinning Gui, Ning Gu, and Peng Zhang. 2021.Data Engagement Reconsidered: A Study of Automatic Stress TrackingTechnology in Use. In
CHI Conference on Human Factors in ComputingSystems (CHI ’21), May 8–13, 2021, Yokohama, Japan.
ACM, New York, NY,USA, 13 pages. https://doi.org/10.1145/3411764.3445763
Permission to make digital or hard copies of all or part of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citationon the first page. Copyrights for components of this work owned by others than theauthor(s) must be honored. Abstracting with credit is permitted. To copy otherwise, orrepublish, to post on servers or to redistribute to lists, requires prior specific permissionand/or a fee. Request permissions from [email protected].
CHI ’21, May 8–13, 2021, Yokohama, Japan © 2021 Copyright held by the owner/author(s). Publication rights licensed to ACM.ACM ISBN 978-1-4503-8096-6/21/05...$15.00https://doi.org/10.1145/3411764.3445763
In today’s fast-paced and hectic world, stress is a growing healthconcern. It is not just that too much stress can reduce study andwork efficiency, but that stress has been closely linked to psycho-logical illness. Previous research has found a strong associationbetween stress and depression [32], and long term exposure to highlevels of stress can negatively impact our well being [15, 66, 80],leading to various physical diseases, such as hypertension [72], car-diovascular disease [72], infectious illnesses [85] and even cancer[21]. Therefore, the awareness and effective management of stressis of significant importance to the management of health.In recent years, stress-tracking technologies that can automat-ically detect and collect stress data during the day have becomecommercially available. More and more wearable products have au-tomatic stress tracking features embedded, such as smart braceletsand watches made by Huawei [39], Garmin [37], and Samsung [40].There are also products that are dedicated to stress-tracking, suchas Healbe Gobe2 [38], Wellbe [42], Bellabeat Leaf Urban [36], andSpire Stone [41]. These products, by automatically detecting anindividual’s stress level, often combined with features to help withrelaxation, offer the potential to help with an awareness and aneffective management of stress on a daily basis.With more mature stress-tracking technologies on the market,however, it is still unclear how these technologies actually work inpractice. In HCI and related fields, there have been many studieson stress, but they’ve primarily focused on innovative approachesto automatic stress tracking [9, 35, 55, 61] or designs that can helpstress relief [20, 70, 92]; little attention has been paid to how peo-ple use the automatic stress-tracking technologies. In addition,although there has been extensive work on the use of self-trackingtechnologies, also known as Personal Informatics (PI) [57], or quan-tified self [17], that are designed to track various aspects of ourlives, such as steps, mood, sleep, and heart rate [13, 33, 54, 59, 78],research on the use of stress-tracking technologies in everydaylives in particular has been rare. Yet, stress-tracking has distinctcharacteristics that deserve their own attention. Stress as a mea-surement is not as straightforward as steps [6], or heart rate thatcan be directly quantified with counting, and is less objective and a r X i v : . [ c s . H C ] J a n HI ’21, May 8–13, 2021, Yokohama, Japan Xianghua and Shuhan, et al. more complex. Furthermore, stress involves not only psychologicaland emotional responses (such as anxiety, anger, sadness [52], fear,and frustration [12]), but also physiological and bodily reactions.As [45] points out, self-tracking for daily stress has unique chal-lenges because stress is highly subjective and involves social andenvironmental factors. Thus, the research question we would liketo answer is this: how do people encounter and use the automaticstress-tracking technologies that have become available in moreand more wearable devices in everyday life?To answer it, we conducted a qualitative study to understandautomatic stress-tracking technology in use. We recruited 17 par-ticipants in China who used automatic stress-tracking technologiesand conducted semi-structured interviews with them. The studyhighlights a number of challenges associated with users’ stress-tracking data engagement, including the lack of immediate aware-ness of relevant data, the lack of pre-required knowledge, domainand technical, as well as the lack of corresponding communitiesof practice. Many of these challenges are associated with how theautomatic stress-tracking technology is adopted and designed, howthe stress data is encountered, and how our users are socially situ-ated. This study unpacks some of the data engagement assumptionsembedded in the related work on self-tracking technologies.The contribution of this paper is an empirical study on the useof automatic stress tracking in practice, and a more nuanced un-derstanding of data engagement with self-tracking technologies.In the paper below, we will first give background information onstress and stress-tracking technologies and review related workson stress and self-tracking data engagement. We will then presentour study and the findings, and discuss how the focus of stress-tracking technologies brings to the fore some of the issues of dataengagement with self-tracking technologies in general.
While the term “stress” is pervasively used, there has never been aunified definition of it. Broadly speaking, stress has been mainlyexamined in two ways, psychologically and physiologically, withthe former focusing on psychological feelings and the perceptionof stress and the latter referring to the bodily response to externalevents.In psychology, stress refers to the feelings and perception ofpressure. It holds that “stress occurs when a person perceives thedemands of an environment stimuli to be greater than their abilityto meet, mitigate, or alter those demands” [51]. Stress is as suchperceived as a subjective concept, and in psychology, self-reportingis usually used to detect it . While most associate stress with nega-tive feelings, such as fear and anxiety, stress can also be positiveand beneficial. Unlike negative stress or “distress”, with positivestress or “eustress” , people appraise a situation to be challengingand non- threatening [26] and have the confidence to solve it. Onestudy found an inverted u-shaped relationship between stress andperformance; in other words, stress is beneficial to performanceuntil an optimal level, and then performance starts to decrease [53].In the medical field, the term “stress” is defined physiologi-cally as “the non-specific responses of the body to any demandfor change” [82]. When people encounter threats or challenges, the body will have corresponding reactions, which are generated bythe autonomic nervous system. The autonomic nervous system(ANS) is comprised of the sympathetic nervous system (SNS) andthe parasympathetic nervous system (PNS). The SNS is responsiblefor mobilizing the body’s resources to deal with stressful events,in what is called the “fight-or-flight” response, and brings with it aseries of physiological reactions, such as an increase in heart rate,and respiration and sweat gland activity [86]. The PNS is mainlyactive during periods of relaxation and recovery.When people talk about stress in everyday life, they are usuallyreferring to psychological or subjective feelings, such as tension,anxiety, and fear, which they frequently associate with an externalevent, like an upcoming exam or deadline at work. While morepeople are starting to realize the impact of long-term stress on theirhealth, the way the nervous system physiologically reacts to stressand the distinction between psychological and physiological stress,are not yet a part of most people’s everyday understanding.
Today, there are wearable commercial stress-tracking devices avail-able on the market that can automatically detect stress includinggeneral products, such as smart bracelets and watches from Huawei[39], Garmin [37] and Samsung [40] with embedded stress -trackingfeatures, as well as other products like Healbe Gobe2 [38], Wellbe[42], Bellabeat Leaf Urban [36], and Spire Stone [41] specializingin stress-tracking. The products by Huawei, Garmin, and Samsungdetect stress based on an analysis of Heart Rate Variability (HRV)collected by an embedded optical heart rate sensor. Some devices,such as the Huawei (including Honor) watch, ask users to fill out astress questionnaire when they first start using the stress-trackingfunction.Heart Rate Variability (HRV), defined as the variation over timeof the period between consecutive heartbeats (R-R intervals) [3],has been proven to be a reliable indicator of ANS activity [64]and can be used as an objective assessment of stress [46]. HRV iswidely used for stress detection with both laboratory stressors (arithmetic problems [29],the Stroop Color Word Test, highly pacedvideo games [14]) and real life stressors (university examinations[68], speeches [5], driving [69]). An accurate HRV is usually ob-tained from an Electrocardiogram (ECG) sensor, which needs toattached by electrodes or chest straps directly to the body. A lessinvasive and more comfortable alternative is Photoplethysmogra-phy (PPG), which can be embedded into a phone camera, ring, orsmart wristband device. The Pulse rate variability (PRV) extractedfrom the PPG has also been proven to be an effective surrogatefor HRV for stress detection [60, 62]. In fact, in many wearablecommercial products, PRV is directly referred to as HRV. In thispaper, we will not distinguish between HRV and PRV , as we willmainly be studying wearable commercial stress-tracking products.These products also commonly provide visualizations to helpusers engage with the quantified stress data. For example, on Huawei’sstress-tracking products, stress value, in range of 1-99, is displayedevery 30 minutes, and is divided into four levels (1-29 as relaxed,30-59 as normal, 60-79 as medium, 80-99 as high), which are shownin bars with different corresponding colours (sky blue, light blue,yellow and orange), as shown in Fig. 1. Unlike Huawei, Garmin ata Engagement Reconsidered: A Study of Automatic Stress Tracking Technology in Use CHI ’21, May 8–13, 2021, Yokohama, Japan shows stress values in real-time. It divides the stress value, in therange of 0-100, into four levels (0-25 as resting, 26-50 as low stress,51-75 as medium stress, 76-100 as high stress); however, except forresting which is represented as blue, all other levels are representedas yellow, and are not further distinguished with different colors,as shown in Fig. 2. After synchronization, users of both Huaweiand Garmin’s products can see more stress details with the corre-sponding mobile applications (Huawei Health, Garmin Connect),including all-day stress data, long-term stress data, the proportionof stress level among other information. These products also pro-vide functions to help users relieve stress, such as deep breathing,biofeedback games, and mindfulness.
In HCI, there are many works on innovative approaches to auto-matic stress-sensing. For instance, office devices such as a mouseand keyboard are explored to detect stress, but this approach is lim-ited to the work environment and cannot be used in other scenariosof daily life [34, 91]. In addition, smartphones embedded with var-ious sensors are commonly studied to detect stress, e.g. based onhuman voices [61], smartphone usage data [90], behavioral metricsof mobile phone activity combined with contextual data [9], andso on. However, they are either too constrained by environmentalfactors (e.g. can’t be too quiet or too noisy)[61] or too limited inaccuracy [90] to be used for daily stress detection.In contrast, wearable devices embedded with sensors are advan-tageous for daily stress detection because they can collect objectivephysiological data and provide relatively timely feedback. For exam-ple, electrocardiograph (ECG) and respiration data obtained from achest belt [35], and pulse rate variability features collected from awatch can all been used to detect stress with a satisfactory accuracyrate in the field, and the wrist device is even more portable and lessinvasive to be used on daily basis.In addition, much of the work is on biofeedback and interventiontechnologies to help people relieve stress. Some explore real-timefeedback on interventions (such as taking a deep breath) to reflecton their behavior patterns [81]. Others compare the relief effects ofdifferent types of interventions including haptic feedback, games,and social networks [70]. Visualizations of stress data with con-texts (such as activity and location) are also explored to inform thecontent and just-in-time interventions [83]. However, it is foundthat the methods of stress feedback need to be carefully designed,otherwise they potentially become stressors [63].Although not specifically focused on stress, in work on mentalwellness, stress management is an important theme. One focus ison improving mental wellness or peacefulness of mind as a wayto deal with stress, e.g through mindfulness [74], or methods tomaintain users’ attention [18, 47, 70]. Some of these studies wereat the intersection of mental health and stress tracking, and somewere design studies, using focus groups [45], or workshops [54],and exploring design opportunities for self-tracking.Overall, the work on stress has mainly focused on innovativeapproaches to automatic stress-sensing and design that can helprelieve stress. Although many stress-tracking products have be-come commercially available, there have been few studies on the use of these technologies in real life. Adams et al. conducted a studycomparing three stress tracking approaches in the real-world envi-ronment (self-report, EDA, and voice-based), and while the studyfound that these three approaches are about equally effective indifferent contexts [4], the study didn’t evaluate automatic stresstracking. In another instance a feature analysis of 26 stress man-agement apps investigated how the apps support reflection andaction [75]; this study also left automatic stress-tracking out fromits analysis. As such, there is still a lack of empirical study andunderstanding of automatic stress-tracking in everyday life.
While little has been done on stress-tracking technology in use inparticular, there has been extensive research on the use of otherautomatic self-tracking technologies, exploring issues of adop-tion/abandonment [19, 23, 76, 79], use in particular domains suchas sports [71] and diabetes management [65], and design for partic-ular tracking [24, 67]. Here we review the self-tracking work thatis related to data engagement.Studies of self-tracking technologies in use reveal various chal-lenges to data engagement. These challenges are often exploredas barriers to adoption, especially for non-experienced users. Forexample, Rapp and Cena focused on how novice users perceiveand use self-tracking tools in everyday life [76], revealing a num-ber of data engagement issues that prevented effective integration,including perceived inaccuracy and untrustworthiness based onthe users’ memories of their behavior, emotional disconnectionfrom abstract visualization, and not knowing what to do with thedata due to a lack of suggestions. Similarly, Lazar et al. found thatparticipants abandoned their devices because they did not think thedata provided anything informative (e.g. when they went to sleepand when they got up), or they did not know what to do with thedata (e.g. what to do with the heart rate data) [50]. Ravichandranet al. ’s[78] study of sleep tracking technology found that users’misunderstanding of what constitutes good sleep restricted themfrom taking meaningful action. In general, it has been found that,except for quantified selfers who are keen on tracking and num-bers [17], most people find it difficult to engage with tracked data[77] for reasons that include incomplete tracking, having too muchor too little data [43], poor aesthetics, unsuitable visualizations, alack of time, a lack of motivation, and a lack of related expertise[33, 50, 57]. As such, users face many challenges when leveragingthe devices’ quantified data for effective use.To support data engagement, visualization and integration havebeen commonly employed to help people gain insights from data,e.g. making the data more ready to use, integrating more contextualinformation, and correlating different sources of data. For example,Li et al. investigated incorporating contextual information to theself-tracked performance data, such as steps, to further promote self-awareness and help people find ways to integrate activity into theirlives [56]. In the study of diabetes self-management practices [65],Mamykina et al. emphasized the importance of a correlation be-tween daily activities (such as exercise, food intake) with the bloodsugar levels to help users reflect and make appropriate lifestylechoices in the future. MONARCA [28] is a system designed for
HI ’21, May 8–13, 2021, Yokohama, Japan Xianghua and Shuhan, et al.
Figure 1: The stress interface on Watch GT Figure 2: The stress interface on Garmin people with bipolar disorder to collect subjective (such as mood,sleep, medicine taken) and objective (such as calls, text messages,physical activity) data through a semi-automatic method that helpsthem identify factors that may affect their disease. Overall, theseapproaches primarily focus on supporting the mental cognitive pro-cesses of data engagement by making data more available, visible,and integrated.In recent years, various novel approaches or designs to supportdata engagement have also been explored. To address users’ chal-lenges, such as low graphical literacy and the inability to uncoversubtle correlations between data sets, approaches beyond visual-ization, including the use of natural language summaries based onstatistical analysis, have been explored to increase data engage-ment and understanding [8, 44]. It has been suggested that somefeatures, like dialogue, might influence the need for reflection [31].Social approaches to reflection have also been investigated. Forinstance, Feustel et al. looked at the idea of aggregating cohort datainto personal informatics systems to support meaningful reflection[25], and Graham et al. conducted a study to understand sharedreflection by asking people to reflect on each other’s data [30].Since the issues surrounding the data engagement of self-trackingis often explored in terms of reflection [57], the conceptual meaningof reflection has also systematically been reviewed and examined tosupport the design for reflection [7, 73, 84]. For instance, drawingon Schon’s notion of reflective practum, Slovak et al. identifiedthree components to scaffold for reflection, explicit (the link be-tween experience and reflection), social and personal components[84]. Baumer reviewed conceptual and theoretical models of re-flection, and identified three dimensions: breakdown, inquiry andtransformation [7]. Ploderer et al. distinguished between reflection-in-action and reflection-on-action, with the former referring to thereflection of realtime feedback, and the latter to data explorationwhen convenient [73]. However, while insightful, these meaningshave evolved from general theories or empirical studies of reflec-tion, and do not involve self-tracking technologies, and thus maymiss the unique complexities and dynamics of reflection broughtby the involvement of the self-tracking technologies themselves.In this paper, we focus on a complex and relatively new andunder-explored automatic tracking technology – stress tracking – inuse, and hope to uncover new insights into users’ data engagement with self-tracking technologies in practice to contribute to this bodyof work and inform the related design of self-tracking technologies.
To gain a better understanding of the use of automatic stress-tracking technologies in practice, we adopted the qualitative re-search method to uncover the rich and detailed usage data, by inter-viewing those who had already used and experienced related tech-nologies. For participant recruitment, we designed a flyer search-ing for those who had used wearable devices with stress-trackingfeatures. On the flyer, we described our study motivation, studymethod, participant qualifications and compensation, and one of theauthors’ WeChat contact information. In China, the two most pop-ular brands of smart wearable devices with stress-tracking featuresare Huawei(including Honor) and Garmin [1, 2], so we posted therecruitment flyers in the QQ and WeChat groups for Huawei andGarmin wearable device users, as well as some general wearablesmart device communities, and our own WeChat circles to recruitmore users. Finally, we recruited 17 participants in total, including14 from the user groups or communities, and 3 from participants’recommendations. Their profile information is shown in Table 1.Most of them are male, and include 16 males and 1 female, largelyaligning with the male to female ratio of smart wearable devicemarket [87]. In addition, most are young, ranging from 21 to 39years old. Their occupations are diverse, and include those relatedto science and technology, such as IT manufacturing trainee, tech-nical developer, programmer, wearable health worker, and exercisephysiology worker, as well as those that are not IT related, includingsalesman, government worker, and customer service employee. Asshown in the table, they were from many different cities in China,ranging from inland cities in the north, such as Beijing, Zhengzhou,Shenyang, and Xi’an, as well as coastal cities in the south, suchas Dongguan, Shenzhen and Guangzhou. Most of the participantsused Huawei’s Watch GT or Honor Watch Magic; P12 and P17 usedGarmin’s Vívosmart 4 and Forerunner 645 Music, respectively. Allhad used the devices from between half a month to 12 months.We then conducted semi-structured interviews with these par-ticipants. Since the participants lived in different cities, most in-terviews were through WeChat voice calls, except for P9, P13, andP14 with whom we did interviews face-to-face. The interviews ata Engagement Reconsidered: A Study of Automatic Stress Tracking Technology in Use CHI ’21, May 8–13, 2021, Yokohama, Japan
Table 1: Participants
ID Gender Age Occupation Device LocationP1 M 21 Exhibition Salesman Watch GT ShanghaiP2 M 31 Government Worker Honor Watch Magic Tonghua * P3 M 21 IT Manufacturing Trainee Honor Watch Magic, Watch GT Dongguan ** P4 M 35 Technical Developer Huawei B5 bracelet BeijingP5 M 22 Design and Operation Worker Honor Watch Magic Zhengzhou * P6 M 28 Programmer Honor Watch Magic BeijingP7 F 26 Wearable Health Worker Watch GT ShenzhenP8 M 31 Business Operator Honor Watch Magic Xuchang * P9 M 39 Government Worker Watch GT ShanghaiP10 M 26 Programmer Honor Watch Magic BeijingP11 M 28 Unemployed Watch GT Shenyang * P12 M 25 Exercise physiology worker Watch GT, Garmin vívosmart4 Finland and ChinaP13 M 34 Customer service Watch GT ShanghaiP14 M 32 Unemployed Honor Watch Magic ShanghaiP15 M 28 Programmer Honor Watch Magic, Watch GT Xi’an * P16 M 24 Advertising designer Watch GT ShanghaiP17 M 24 Student Garmin Forerunner 645 Music Guangzhou * inland cities in northern China ** coastal cities in southern China usually lasted about 40 minutes. During the interviews, we askedparticipants for basic information including their age, occupation,location, and education, as well as questions of their general useof devices and how they experience and manage stress, such aswhat devices they had used, what applications in the devices theyused most frequently, how they used these applications, their stressstatus, their perceived stress source(s), and how they deal withstress on a daily basis. We then asked about details of their use ofstress-tracking technology, probing for concrete usage instances,e.g under what circumstances they checked the stress data andhow, what they saw and how they experienced and understood it,what they did after seeing the data, etc. For some usage instances,we asked whether they could provide screen shots of their stressapplication interfaces for clarification, and some sent screen shotsover WeChat to us. We followed up with some of the participantsafter the interviews, keeping in touch through WeChat to knowmore about their use of the devices, special events they found intheir later use, and changes in their long-term stress status. We alsocollected these chats for later data analysis.All interviews were conducted in Chinese Mandarin. With theconsent of the participants, we audio-recorded the interview pro-cess and transcribed it into text for later data analysis. For privacypurposes, we anonymized their data in the transcript and in thepaper.We conducted thematic analysis inductively [11] with the in-terview data. We first familiarized ourselves individually with thedata and then extensively read, analyzed, and discussed it together.Each of us generated our own set of codes, and we compared ourcodes in meetings and discussed it further. We eventually identifiedthe challenges of engaging and understanding data as the primarytheme. We identified three sub-themes under this key theme, whichwe report in our findings. For privacy purposes, we anonymized our interview in the paper by using P Almost all of our interview participants had already adopted andintegrated their smart wearable devices into their everyday lives,so we do not have adoption issues as discussed in previous works(e.g.[76]). Our participants wore their watch or bracelet all the timeexcept for occasions when it was not feasible, such as when it wascharging or they were taking a shower. All of our participants,except P2 and P10, had not bought the watch or bracelet for theprimary purpose of stress-tracking, but for other reasons includingsports, to not miss phone calls, or to replace a traditional watch. Infact, most were not aware of the existence of the stress-trackingfeature at the time of purchase and only discovered it later whileexploring the device. P17, who loved jogging, offered a typicalexplanation: “
I didn’t even know there was a stress-tracking feature.I didn’t buy the Garmin watch for stress detection. I bought and used[the watch]. It was after I used it that I got to know this feature. ”Although participants did not adopt the technology for the sake oftracking stress, they all quickly became aware of the feature as it isquite accessible, just a few clicks or swipes away.We also found that, while participants could easily access thestress data, their understanding of it was quite varied. Only a fewcould meaningfully engage with it, some only had a limited under-standing, while others were confused. At the same time, severalparticipants reported that just the awareness of the existence ofthe stress tracking feature, not necessarily an understanding ofthe data, had some impact on their behavior, as similar to whatis found in [27]. For P13, simply being aware that the watch con-stantly monitored his stress helped him to watch his temper: “
Soat that time when I did not have the watch, I would not deliberately
HI ’21, May 8–13, 2021, Yokohama, Japan Xianghua and Shuhan, et al. control my mood or emotions, and I would just discharge. After wear-ing this watch, to some extent, I felt that I was monitored every day,so I couldn’t make myself too stressed, or lose my temper. ” Similarly,to P10, the awareness of the feature had impact on his behaviorsubconsciously: “
If you have [the stress tracking feature], you willsubconsciously adjust yourself... ” However, not being able to effec-tively engage with the stress data overall kept most of them frommaking more informed use of it. Below, we turn our focus to thechallenges of stress-tracking data engagement we uncovered fromthe study.
All of our participants were excited about the automatic stress-tracking feature when they began using their devices. They checkedthe data frequently. However, after one or two months, the noveltyeffect was gone, and the majority (except P12) stopped engagingwith the data in a timely and frequent manner. P2’s situation istypical among our participants: “
I paid close attention to the stressdata when I just started wearing the watch. I checked the data at leastten times a day, but I checked it less and less frequently over time. It’sbeen 2 months since I started using it, and now I only check the datatwice per day. ”The main reason for this was that our participants did not feelthat the stress-tracking devices were helpful to raising their imme-diate awareness due to their natural responses to some stressfulevents and the limitations of the devices. First, when people en-counter challenges, their mind is usually too occupied by distressingthoughts [22] to break away from their minds and check the stress-tracking data. Participants (except P12) commonly reported thatwhen they were facing challenging issues, they were overwhelmedand would intuitively focus on solving their issues rather thanchecking the stress data. For instance, P3 told of a time when he gotassigned a challenging task at work that made him feel so stressedthat instead of checking his stress level, he focused on the taskfirst: “
I was assigned with some tasks last time, when I just startedmy internship. wow, what the hell! I could not believe it. I then askedothers a lot of questions and looked up information. I was very stressed... But when you are truly stressed, surely you don’t think of looking atthe watch – you think of solving the current problem first. ” Similarly,P5 noted: “
When I am busy and stressed, to be honest, I don’t paymuch attention to [the data]. I would probably just check the time onthe watch at best. ”In rare cases, when participants were stressed to the point ofphysical discomfort, however, they might look at the data in thatmoment. As P2 reported: “
When I felt that my heart was beating abit fast, I would take a look at it. Or sometimes, when I was writing alot of summaries or textual materials, and felt dizzy, I would take alook at it and pay more attention to stress. ” However, in these cases,devices that don’t provide real-time updates made it difficult to gainan immediate awareness. For example, Huawei’s stress-trackingapp only updates every half hour, and P12 reported how it causedconfusion sometimes as the data was not consistent with what theyfelt in the moment, “
You can feel your own heartbeat rising. You calla customer, or do an interview, and you feel your heartbeat rising,but when you look at your watch, the stress data has not risen. ” Ourparticipants complained how Huawei’s delay discouraged them from checking it in the moment they felt their emotions intensify.P3 said: “
The problem is... I think it should update more frequently...If your stress rises again within half an hour, it can’t be monitoredat all... ” As such, this lack of timely feedback makes it even moredifficult for people to develop an immediate awareness of theirstress status.Considering it is challenging for people to remember to usetheir stress-tracking devices and the importance of in-the-momentinterpretation, it is critical that stress-tracking devices help raiseimmediate awareness. However, our participants reported that thedevices failed to notify them when their stress levels were highor changed remarkably. This made our participants feel like thedevices were not helpful. For example, P4 explained,“
I now feel this function is not so meaningful. . . since itdoesn’t react or intervene in time. I often only realizethat I was stressed out after my stress has gone. . . Suchdevices should help people manage real-time stress ratherthan just recording it, right?...Only recording it is notso helpful. I was interested in checking the data in thebeginning because it seemed to be a novel function, butnow, since it’s not so helpful, I don’t check it frequentlyanymore. ”Even P12, who was the most engaged user among all of our partici-pants, said,“
Both devices (Huawei Watch Gt and Garmin watch)claimed they have something like a stress level reminder,but I have never been reminded...even when my stresslevel was as high as between 80% to 90% in Garmin...Iwish they could provide an in-the-moment reminder andprovide us effective, timely ways for stress management,such as providing relaxing music for stress relief...If theycould remind us that our stress was high in the moment,we would have more interactions with the data andcould manage our stress better. . . ”Users’ in-the-moment data engagement is critical for reflection [84]and intervention. Failing to provide timely reminders hinders usersfrom effectively managing their stress.Most of the time, they noticed the stress data through a casualor random encounter with the technology. That is, they did notintentionally check the data; rather, they only noticed it when theywere browsing other types of data (e.g., heart rates) on their devices,when casually playing with their devices when bored, or when justtaking a glance at the devices while taking them off. During thesecasual encounters with the data, something would stand out andbecome noticeable, drawing their attention to it. Most commonlythis was a sudden rise in stress level or a change of color. Forinstance, P10 once noticed an unexpected, sudden rise in his stresslevel after lunch after having a period of relative stability: “
I usuallytake a nap during the lunch break, and usually my stress is quitestable. However, there was one time, after having a nap, I saw a suddenrise of stress level when I was randomly playing with my watch. Thestress was quite high. ” P5’s stress data caught his attention due toits change of color: “
Yeah, on the 13th of this month, just two daysago. . . I saw the yellow color for the first time. The colors range fromblue to green to yellow to red. Yellow means the stress is quite high. Itwas the first time for me to see such a high value. ” As shown in these ata Engagement Reconsidered: A Study of Automatic Stress Tracking Technology in Use CHI ’21, May 8–13, 2021, Yokohama, Japan cases, the visual presentation of the data (e.g. the bright color overa dark background, the sudden change) against the participants’personal experiences (the first time seeing it) led them notice it.As a result, while the automatic stress-tracking devices afford richdata, only a very small portion of it actually drew our participants’attention and motivated them to interpret or reflect on it.As these cases also illustrate, not all quantified numbers receivethe same attention, or have equal importance – only some of theinformation stands out and matters. Data presentation or visualdesign plays a certain role here in filtering out this information andtelling users where to draw their attention. However, the partici-pants’ attention was usually only drawn to the data during casualencounters. Overall, our participants did not engage with the datafrequently because the devices failed to raise their immediate aware-ness and provide effective in-the-moment interventions.
When some of the data actually drew our participants’ attentionand motivated them to interpret or reflect on it, most of our par-ticipants (except P12) found it challenging to make meaning outof it. This is primarily because our participants had adopted thepsychological notion of stress, while the devices measured phys-iological stress. In other words, our participants perceived stressas a subjective concept that refers to the feeling and perceptionof pressure while the automatic stress-tracking devices measuredstressed physiologically by analyzing bodily reactions generated bythe autonomic nervous system. Unlike tracked activities, which aremore straightforward for quantification and interpretation, suchas steps and hours of sleep, measures such as stress are often morechallenging for interpretation. The distinction between the psy-chological notion of stress that our participants adopted and thephysiological notion of stress that the devices were based on ledto barriers that prevented our participants from interpreting andmaking meaningful use of the tracked stress data.Among our participants, only P12 understood that the deviceswere based on HRV and measured stress physiologically ratherthan psychologically. He managed to learn the related technical anddomain knowledge early on by searching online and reading relatedscientific articles: “
I read some articles on the Internet, and somepopular science articles about what the autonomic nervous systemis and the relationship between the heart rate variability and theautonomic nervous system. ” So he had the basic understanding thatthe stress measured by the devices corresponded to how his bodyresponded to external demands. However, our other participantswere confused when interpreting the data, since it didn’t reflect theirsubjective feelings of stress, that is, their perceptions of pressure.In everyday conversation, when people say “stress,” it usuallymeans psychological stress. Asking to fill out a stress questionnaireto use this feature on some products such as Huawei’s furtherreinforced such a perception. Thus, our participants (except P12)felt that the device wasn’t helpful when they discovered that themeasurements from the tracking technology did not match theirfeelings. When triggered by “feeling something”, our participantsoften expected to see that reflected on the tracking technologyand became disappointed when it wasn’t. P11 explained: “
I hadjust bought it, and at that period of time I was under great pressure; however, it did not show it when I felt stressed several times. I can’tremember what happened exactly. I just remember that I specificallylooked at it when I’d just bought it and felt stressed but it didn’t changeas much as I’d imagine. ” P13 had a similar experience and thoughtthat the stress-tracking application was inaccurate: “
I found it wasinaccurate when I began to use it. I was unhappy and lost my temperat that time, and I found (my measured stress was) just ’medium’instead of ’high’. ”Moreover, some participants found that there was often a correla-tion between their physical activities, such as eating and exercises,and changes in their stress levels on the device, which baffled them.P1 noticed that his detected stress level rose after lunch: “
I don’tthink it’s accurate...Most time it is around noon, such as after eat-ing, the stress is higher. I don’t understand why... ” Only after weexplained that the device was based on HRV and that eating couldput a physiological burden on the body because of digestion, didhe think it made sense. Similarly, P6 thought that a lot of things inhis life stressed him out and was puzzled why he did not see thembeing manifested in the application. He wondered what counted asstress: “
How could there be no stress in my life? I need to paya mortgage monthly, which is definitely stressful. I’mquite worried every time I think about it. [Yet] This isn’treflected [on the watch]. Doesn’t the situation count asstress? I don’t understand. It is not reflected anyway. Ifit could be reflected, I think the watch would show mystress level as yellow everyday. ”As such, the complexity of stress – e.g. involving both psychologicaland physiological, the external and the internal – makes interpret-ing its data more difficult than other tracked data, such as steps andcalories, on the same device.Oftentimes, the displayed stress range (e.g. “relax”, “normal”,“medium stress” or “high stress”) did not fall into our participants’subjectively felt understanding. For example, P2, whose stress levelhad been high on the application, was doubtful as to its accuracy,as the application never displayed a “low” level, even when he wasengaging in relaxing activities: “ . . . for instance, when I go out towatch a movie, eat, or have fun, like sing karaoke with friends, ofcourse I don’t have stress, but the watch still showed that my stress washigh.... ” P2 even tried to recalibrate the measurement by retakingthe stress questionnaire:“
I always doubted that whether that’s because my an-swers to the questionnaire were too pessimistic, and thewatch thus set the baseline stress scale higher than howI truly felt. Thus, I unbonded it with my account. . . andretook the questionnaire. I intentionally answered thequestions more positively than the first time. It turnedout that the measured stress levels overall have indeeddecreased a bit, but are still higher than how I feel. I’mconfused. Why is it always high, whether I am relaxedor not? ”When evaluating whether the stress data was accurate or not, P2was comparing the data against how he felt, i.e, his psychologicalstress level. The mismatch between the physiological type of stressthat the device measured and the psychological stress that P2 per-ceived led to P2’s confusion. Similar to what is found in previous
HI ’21, May 8–13, 2021, Yokohama, Japan Xianghua and Shuhan, et al. work [23, 50], this kind of perceived inconsistency caused our par-ticipants to distrust the system and even led some to stop payingattention to it.To make things worse, the underlying technological mechanismof the application was not so straightforward either. In our inter-view study, many expressed that they did not know how the stresswas sensed by the devices. They had different kinds of speculations.For instance, P1 asked us, “
Is it based on some kind of algorithmsto calculate my stress level? Or is it monitoring my blood pressuresor something through my skin? ” Some assumed it corresponded totheir real-time heart rate. P7’s interpretation was typical: “
My un-derstanding is that it mainly depends on your heart rate...For example,if your heart rate is relatively high, it will recognize that you maybe a bit more stressed. ” P3 also guessed that the measured stresslevel was determined by the heart rate, noting “
If the stress is higher,the heartbeat will be faster. ” However, this theory soon led to fur-ther confusion, as participants discovered it was not exactly right:“
In the afternoon I went to other places and took a look. I had a lotof activities, but the stress was not high. I don’t think the stress isbased on the heart rate. ” P5 went through a similar process whenhe realized: “ My heart rate was high during exercise, [but] the stressvalue was normal. ” Almost all our participants wanted to learn moreabout the underlying mechanism behind the stress-tracking. Forinstance, P17, when asked whether there was anything he did notunderstand, explicitly told us that he had agreed to be interviewedbecause he wanted to find out how the application detected stress:“
How is it measured? It should be calculated by some algorithm, but Idon’t know what specific algorithm it is... I’m definitely curious. Thatis why I accepted [your interview], because I’m curious. ”Additionally, the conditions for stress-tracking also caused moreconfusion, as the devices only sensed stress when one was still. P12reported such confusion: “
In the beginning, I didn’t know why mystress value was not shown at noon. It was weird. You see, the stressvalue usually disappeared from 11:00 to 12:00, and it came out againfrom 12:00 to 13:00. It was strange. Why did it usually disappear at11 o’clock? ” Only after we explained to him, did he realize it wasbecause he was actively moving around noon that stress data wasnot detected. In a word, the not so intuitive mechanism for stresstracking led more confusions.In summary, even when our participants paid attention to andtried to interpret the data, it was challenging for them to decipherit due to the mismatch between what the devices measured andwhat our participants considered to be stress, as well as the un-clear underlying technological mechanisms of the automatic stresstracking.
Despite that our participants encountered challenges when tryingto make sense of the stress data, they did not have easy access torelated resources and communal support to help them tackle thechallenges to achieve meaningful interpretations.Alone among our participants, P12, who lives in both China andFinland, was primarily working in Finland at the time, represents acontrasting case. When working in Finland, he was situated in asocial world which helped him develop a shared understanding. Hedescribed, “
In our company, quite some employees are wearing sports bracelets. I feel 60 percent are wearing these, so it is also part of ourtopic, and we chat quite much about it...[stress tracking] is a functionof our lives and is something we all use. ” The socialization at P12’scompany also helped him to understand the application better. Heexplained, “
I had a cup of coffee when getting up in the morningand I felt relaxed, but my watch showed my stress was medium. So Iasked my colleagues, ‘Is it because of the coffee?’ and my colleaguessaid it was... ” Without understanding that stress is not simply apsychological concept but also a physiological one, it might notbe so easy to see that drinking coffee can cause stress levels torise; this would be even more difficult for someone to process ifthey actually felt relaxed after doing so. More studies have shownthe importance of socialization or social processes for learningand forming shared background understanding for interpretation(e.g. [49]). P12’s experience is a case in point. Being part of such acommunity provides the social means to acquire related knowledgeand collectively address matters of confusion.Unfortunately, for the other participants who were all in China,corresponding stress-oriented communities of practice still havenot yet formed to help develop the need for shared understanding.In China, while sports-related tracking technologies have becomequite popular, and many social groups have formed, stress-trackingis still new and something of which people are rarely aware. How-ever, P17 provides a nice example that illustrates how being partof a community of practice can make a difference. P17, who joinedseveral sports-related groups and bought the watch for sports asmost of members in the group did, reported how he could easilyinterpret and meaningfully read the tracked numbers for pace andheart rate:“
I just check to see if my pace matches my heart rate. Ifyou know how to run, then (you will know), for example,if I run at a pace of 6, then my heart rate should beabout 140. If I run at a pace of 6 one day but my heartrate suddenly reaches 150, I will know that my athleticability has dropped; if my pace is 6, and my heart ratebecomes lower, e.g. it was lower than the previous 140and was 130, I will know that I have improved. ”By looking his pace and heart rate, he could easily tell whetherhis athletic ability had improved or not. On the other hand, ashe explained, it was not easy for him to interpret the stress data:“
For those who bought the watch just for running, they would notunderstand it at all. He may only understand that the Chinese wordsor number there shows something about stress, but he wouldn’t knowwhat exactly the words or the number means. There is no way for meto know, and I believe most people wouldn’t know either. I only havea vague knowledge about it. ” As such, while he could meaningfullyread the running- related numbers, the stress numbers still puzzledhim.In other words, the broader social and cultural context shapeshow people approach the tracked stress data and whether theyunderstand it. P12 described the different accessibility of relatedlearning resources about sports and about stress in China today:“
For example, there are many books about running on the Internet inChina, but there have not been books really about life stress...In China,I think there is still a lack of knowledge about stress management orlife management. ” P12’s observation was confirmed by our other ata Engagement Reconsidered: A Study of Automatic Stress Tracking Technology in Use CHI ’21, May 8–13, 2021, Yokohama, Japan participants’ experiences. P2, who adopted the watch to learn moreabout his stress, only knew that the watch confirmed that his stresslevel was generally high, but did not know how to deal with it.Some participants reported that they only wanted to know whethertheir stress level was normal or not, but they were usually unableto tell by just looking at the data. This is to say, due to the lack ofsupport from the broader socio-cultural context in China, it is noteasy for people to develop a meaningful reading of the data beyondwhether or not their stress level is high or low.
In the preceding sections, we presented a study on the use of theautomatic stress tracking technology in practice, highlighting threechallenges presented by its data engagement: a lack of immediateawareness preventing engagement with data in-the-moment, alack of pre-required knowledge, domain and technical, and a lackof communal support. As shown here, these challenges are notmerely associated with one’s capabilities, such as graphical literacyor quantitative analysis capabilities as pointed out in prior works[76], but have to do with factors embedded in corresponding socialpractices of stress tracking with the technology, such as people toooccupied to check the stress data in the moment when they werestressed, the technology failing to provide timely feedback andreminders, as well as the mismatch between the scientific notion ofstress and the everyday use of stress. While previous works revealedsimilar challenges, such as a lack of the expertise needed to interpretthe tracking data [50], these challenges were mainly identifiedas reasons for adoption or abandonment. Data engagement itself,and its association with corresponding social practices, have notreceived sufficient attention. Drawing on our study of the use ofthe automatic stress tracking technology in particular – a relativelyrecent development and more complicated technology, we madedata engagement our focal point, and unpacked the underlyingreasons contributing to these challenges.
As shown in our study, one challenge of stress-tracking data engage-ment comes from how the data is encountered . Automatic trackinghas been primarily explored to relieve the burden of manual track-ing [16]. However, just as Choe [17] and others (e.g. [58]) havepointed out, automatic tracking can reduce engagement and aware-ness, which can also compromise its effects. More importantly, asrevealed in our study, the users’ intention to track the data, that isoften assumed in self-tracking work, is simply not there anymorefor the stress tracking feature coming with the wearable device. Butrather, they noticed their stress data while engaged in their dailynon-stress-oriented practices. This is quite different from those whodeliberately engage in an active management practice, with thetracking technology intentionally employed to facilitate a process,e.g. to help address their insomnia issues [78] or manage a chronicillness such as diabetes [65]. We can label this casual encountermode to distinguish it from serious encounter mode in which usershave the intention to track and engage with data.When the users’ intention to track data can simply be assumed,we can reduce self-tracking data engagement to a mere analytic issue, as has often been the approach in prior work for data engage-ment [28, 56, 65]. In the casual encounter mode, however, users justencounter the data being collected and presented to them. Insteadof pulling the data out for analysis themselves, users are pulledtowards certain data that draws their attention. As a result, onlya small amount of data will become “present” to them, and whatdata becomes present is highly dependent on how it is presentedand the interactive practices users use to engage with the devices.That is, these two different modes, casual encounter and serious en-counter, are totally different in terms how and what data will cometo the users’ conscious attention for engagement. With more andmore automatic self-tracking technologies embedded in everydayobjects such as smart watches, we believe the non-intentional andcasual encounter will become increasingly more commonplace. Toaddress the data engagement issues in casual mode then, we shouldgo beyond simply user interface revisions which would not makethese devices work, and consider the practice as a whole. Below wediscuss implications based on the casual encounter mode and thechallenges we identified from the study.
To support casual users engaging with the automatically collecteddata, there needs to be not simply new ways of data presentationor integration, but new interactive designs to help engage users inthe right moment. For example, in addition to increasing trackingfrequency, we, like many of our participants, think it would be valu-able to provide reminder functions at appropriate times, e.g. whenthere is a big jump or drop in stress level, or when the stress levelexceeds a certain threshold, in order to fully support the situatedinterpretation process. The reminder functions could help userscapture, engage with, and interpret the notable data in the momentwithin the situated context, not from hindsight. Of course, the re-minder should be provided in a peripheral and non-intrusive way,such as a vibration. Design studies should be conducted to identifywhat moments will be good for sending alerts and how. Also, usersshould be allowed to customize whether to turn on the alert andunder what circumstances need the alert. What we would like toemphasize here is that this is different from integrating contex-tual information into the data presentation as is often approached,but is an approach that supports situated data engagement and“reflection-in-action” [84] through which one’s experience could bedirectly drawn on for data interpretation and correlations.
Another challenge comes from the kind of expertise needed fordata engagement. As shown in the study, the meaningful interpre-tation of the stress data requires necessary knowledge, includingthe domain knowledge of stress and the technical knowledge ofthe mechanism of stress monitoring. Intentional tracking and anactive management practice often mean that users already haveacquired the requisite knowledge needed for the use of the self-tracking technology, e.g. knowing what glucose or blood sugarlevel means and how they should be managed. However, with moredata being automatically tracked and readily available with thewearable devices, as in our case, this prerequisite can no longer
HI ’21, May 8–13, 2021, Yokohama, Japan Xianghua and Shuhan, et al. be assumed. Those who sought out the relevant resources to gainrelated expertise on the Internet, such as popular science articleson stress or videos related to how to use the Garmin watch, wereable to overcome some of the challenges. However, for most users,despite their interest and curiosity, there was no easy access tolearning resources.Besides automatic tracking, this challenge also becomes moresalient with the nature of stress itself, a physiological state thatis more complex and less straightforward than other measures,such as steps. A close examination of our data suggests that some-times the users’ perceived inaccuracy was often due to a mismatchbetween their subjective experiences and the qualitative presenta-tion of the stress data. For example, when they experienced stress,users expected to see their tracked stress presentation to be “high”and match their subjective feelings, rather than “medium” or “low,”which made them perceive the technology as inaccurate; althoughthe trend of the change in quantitative terms, the rising or droppingof the curve, actually corresponded well with their change of feel-ings. As such, dealing with health or physiological states similar tostress, things that can only be experienced subjectively, could causemore trust issues and add more challenges to data engagement.A meaningful reading of this kind of health data, such as stress,heart rate and sleep, thus requires more specialized knowledge andcan pose more challenges for lay people. However, as the majorityof the work of automatic self-tracking technologies thus far hasfocused on relatively more straightforward data, these challengeshave not been sufficiently emphasized.Our study also revealed that different layers of understandingcan be achieved through different levels of acquired knowledge;while some simply got a clearer idea of their stress level, others,such as P12, developed a more meaningful understanding of howstress correlated to different aspects of their lives. We can call theformer “direct understanding” and the latter “deepened understand-ing”. The difference between the two is similar to the differencebetween a lay person and an experienced doctor who can reada lab test result. While the lay person can only tell whether theresults are normal or not (i.e., within normative range), the doctorcan tell whether the patient’s condition has improved and whetherthe immune system has become stronger. To achieve a “deepenedunderstanding”, a deliberate effort to acquire related knowledgeand expertise is needed.Therefore, for the automatically tracked health data, it is crucialto make the requisite knowledge more available and present it ina meaningful way to help users interpret the data. In the case ofstress-tracking, it means making it more possible for users to learnmore knowledge about stress, more about what stress means in thedevice, and more about what the mechanism for tracking is, amongother things. For a more meaningful engagement and deepenedunderstanding, design that helps people acquire related expertisebecomes even more important; we could consider integrating learn-ing materials, in terms of short texts, pictures, or videos, into theproducts to make them more accessible and to present them in amore compelling way. More work is needed here to specificallyunderstand how knowledge could be presented for users to inter-pret their data more effectively, e.g. integrating relevant knowledgethrough visualization, leveraging intelligent conversation agentsto support user inquiry about the data, etc.
For meaningful data engagement, it is also important to supportthe social processes of "knowing." What we have seen in the studyis that the concept of stress in science still has not been mergedwith the concept of stress in everyday life [88], and methods forstress-tracking and stress management have still not become partof popular culture, adding to the difficulty of interpreting trackedstress data. P7’s case is the telling one. While he could easily in-terpret the sports data and gained a corresponding understandingfrom the sports community of practice, the stress data was stillpuzzling to him. Contemporary anthropological and sociologicaltheorizing has already illustrated that participation in the socialworld is a fundamental form of human learning/knowing [49]. Laveand Wenger, for instance, focus on social engagement and partic-ipation as the context in which learning occurs [49], and call thebroader context, or the social world, “communities of practice”. Itis through communities of practice that resources are shared, in-formation spreads, and shared understanding is achieved. Whereself-tracking technologies are concerned, we believe participationin corresponding communities of practice is the key to go fromsimple tracking to "knowing."Recognizing the social nature of learning/knowing provides adifferent perspective for design, e.g. facilitating the forming ofcorresponding communities of practice is as important as makinglearning resources easily accessible. Think about the recent devel-opment of open science [89] or citizen participation in the scientificinquiry processes; this provides a valuable model not simply forscientific discovery, but also for scientific education [10], or themerging of scientific and everyday knowing. This is different fromsocial discussions for data analysis [25, 30], and is more of a com-munity that helps members learn through their participation andsocial interactions. Supporting the formation of communities ofpractices and user participation, could be an effective approach tothe support of learning and "knowing" of self-tracking technologies.
We note that although the gender ratio of our participants largelyaligns with the gender ratio of smart wearable device users in China[87], there is only one female participant in our study, which mighthave introduced gender bias. For future work, with automatic stress-tracking features becoming more available in wearable devices, itwould be helpful to have more studies of this kind look into detailedusage across different sites and diverse populations. Our study alsosuggests promising directions for future design explorations, mainlyto address the data engagement challenges identified in this work,e.g. mechanisms to support in-situ data interpretation, effectiveways to integrate corresponding expertise knowledge with the data,and ways to develop communal support to help users form a sharedunderstanding.
In this paper, we present a study of the use of a relatively recent andless straightforward stress-tracking technology-in-practice, high-lighting the three primary challenges of data engagement withautomatically-tracked stress data: a lack of immediate awareness, a ata Engagement Reconsidered: A Study of Automatic Stress Tracking Technology in Use CHI ’21, May 8–13, 2021, Yokohama, Japan lack of prerequisite expertise, and a lack of corresponding communi-ties of practice. In particular, by focusing on a relatively “unfamiliar”stress-tracking technology, the study reveals that many assump-tions that have been taken for granted about self-tracking technolo-gies no longer hold true in the rapidly changing world. Reflecting onthe challenges uncovered from our study as well as related works,it is clear that some elements of the technically-mediated trackingpractices make a difference as far as data engagement is concerned.These include how the data is tracked /encountered – intentionallyor unintentionally; who the users are – novice or expert; what istrack – activity or health states; and where the users are situated –in a community of practice of corresponding tracking or not.With the development of self-tracking technologies, and withincreasingly more automatically-collected health data made easilyavailable in our lives, it does not simply help to reduce the laborneeded for tracking [17], but rather more fundamentally, to changethe very mode of data engagement in practice. What we highlightthrough the study is that the meaning of self-tracking is not simplya matter of having the data, nor analyzing the data, but a matter ofsituated practices of data engagement. As shown in the study, farfrom being simply an interaction between users and their data, dataengagement is embedded in a web of an individual’s intention fortracking, domain knowledge, technical properties, other users, andlearning resources. As Kuutti put it: “Practices are wholes, whoseexistence is dependent on the temporal interconnection of all theseelements, and cannot be reduced to, or explained by, any one singleelement” [48]. We argue that to understand the data engagementissues of self-tracking technologies, we should approach them aspart of the whole tracking practice, a notion that assembles all theseelements into a holistic unit, and does not reduce them to a merecognitive analysis.
We would like to thank our participants for sharing their experi-ences. This work is supported by the National Key Research andDevelopment Plan of China award number(s): 2016YFB1001200,the National Natural Science Foundation of China (NSFC) awardnumber(s): 61672167, 61932007.
REFERENCES [1] [n.d.]. Share of Chinese smart bracelet brands. http://top.zol.com.cn/compositor/trend_829.html[2] [n.d.]. Share of Chinese smartwatch brands. http://top.zol.com.cn/compositor/trend_827.html[3] U Rajendra Acharya, K Paul Joseph, Natarajan Kannathal, Choo Min Lim, andJasjit S Suri. 2006. Heart rate variability: a review.
Medical and biological engi-neering and computing
44, 12 (2006), 1031–1051.[4] Phil Adams, Mashfiqui Rabbi, Tauhidur Rahman, Mark Matthews, Amy Voida,Geri Gay, Tanzeem Choudhury, and Stephen Voida. 2014. Towards personalstress informatics: comparing minimally invasive techniques for measuring dailystress in the wild. In
Proceedings of the 8th International Conference on PervasiveComputing Technologies for Healthcare . ICST (Institute for Computer Sciences,Social-Informatics and ..., 72–79.[5] Ana C Aguiar, Mariana Kaiseler, Hugo Meinedo, Traian E Abrudan, and Pedro RAlmeida. 2013. Speech stress assessment using physiological and psychologicalmeasures. In
Proceedings of the 2013 ACM conference on Pervasive and ubiquitouscomputing adjunct publication . ACM, 921–930.[6] David R Bassett, Lindsay P Toth, Samuel R LaMunion, and Scott E Crouter.2017. Step counting: a review of measurement considerations and health-relatedapplications.
Sports Medicine
47, 7 (2017), 1303–1315.[7] Eric PS Baumer. 2015. Reflective informatics: conceptual dimensions for designingtechnologies of reflection. In
Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems . 585–594.[8] Frank Bentley, Konrad Tollmar, Peter Stephenson, Laura Levy, Brian Jones, ScottRobertson, Ed Price, Richard Catrambone, and Jeff Wilson. 2013. Health Mashups:Presenting statistical patterns between wellbeing data and context in naturallanguage to promote behavior change.
ACM Transactions on Computer-HumanInteraction (TOCHI)
20, 5 (2013), 30.[9] Andrey Bogomolov, Bruno Lepri, Michela Ferron, Fabio Pianesi, and Alex SandyPentland. 2014. Daily stress recognition from mobile phone data, weather condi-tions and individual traits. In
Proceedings of the 22nd ACM international conferenceon Multimedia . ACM, 477–486.[10] Rick Bonney, Tina B Phillips, Heidi L Ballard, and Jody W Enck. 2016. Can citizenscience enhance public understanding of science?
Public Understanding of Science
25, 1 (2016), 2–16.[11] Virginia Braun and Victoria Clarke. 2019. Reflecting on reflexive thematic analysis.
Qualitative Research in Sport, Exercise and Health
11, 4 (2019), 589–597.[12] Gillian Butler. 1993. Definitions of stress.
Occasional paper (Royal College ofGeneral Practitioners)
61 (1993), 1.[13] Clara Caldeira, Yu Chen, Lesley Chan, Vivian Pham, Yunan Chen, and Kai Zheng.2017. Mobile apps for mood tracking: an analysis of features and user reviews. In
AMIA Annual Symposium Proceedings , Vol. 2017. American Medical InformaticsAssociation, 495.[14] Rossana Castaldo, Luis Montesinos, Paolo Melillo, Sebastiano Massaro, and Lean-dro Pecchia. 2017. To What Extent Can We Shorten HRV Analysis in WearableSensing? A Case Study on Mental Stress Detection. In
EMBEC & NBC 2017 .Springer, 643–646.[15] Tarani Chandola, Eric Brunner, and Michael Marmot. 2006. Chronic stress atwork and the metabolic syndrome: prospective study.
Bmj
IEEE Pervasive Computing
16, 1 (2017), 74–84.[17] Eun Kyoung Choe, Nicole B Lee, Bongshin Lee, Wanda Pratt, and Julie A Kientz.2014. Understanding quantified-selfers’ practices in collecting and exploringpersonal data. In
Proceedings of the SIGCHI Conference on Human Factors inComputing Systems . ACM, 1143–1152.[18] Michael E Clark and Richard Hirschman. 1990. Effects of paced respiration onanxiety reduction in a clinical population.
Biofeedback and Self-regulation
15, 3(1990), 273–284.[19] James Clawson, Jessica A Pater, Andrew D Miller, Elizabeth D Mynatt, andLena Mamykina. 2015. No longer wearing: investigating the abandonment ofpersonal health-tracking technologies on craigslist. In
Proceedings of the 2015ACM International Joint Conference on Pervasive and Ubiquitous Computing . ACM,647–658.[20] Karen Anne Cochrane. 2019. Reconnecting the Body and the Mind: Technology toSupport Mindfulness for Stress. In
Extended Abstracts of the 2019 CHI Conferenceon Human Factors in Computing Systems . ACM, DC05.[21] Sheldon Cohen, Denise Janicki-Deverts, and Gregory E Miller. 2007. Psychologicalstress and disease.
Jama
Stress management for dummies . John Wiley & Sons.[23] Daniel A Epstein, Monica Caraway, Chuck Johnston, An Ping, James Fogarty,and Sean A Munson. 2016. Beyond abandonment to next steps: understandingand designing for life after personal informatics tool use. In
Proceedings of the2016 CHI Conference on Human Factors in Computing Systems . ACM, 1109–1113.[24] Daniel A Epstein, Nicole B Lee, Jennifer H Kang, Elena Agapie, Jessica Schroeder,Laura R Pina, James Fogarty, Julie A Kientz, and Sean Munson. 2017. Examin-ing menstrual tracking to inform the design of personal informatics tools. In
Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems .6876–6888.[25] Clayton Feustel, Shyamak Aggarwal, Bongshin Lee, and Lauren Wilcox. 2018.People Like Me: Designing for Reflection on Aggregate Cohort Data in PersonalInformatics Systems.
Proceedings of the ACM on Interactive, Mobile, Wearable andUbiquitous Technologies
2, 3 (2018), 1–21.[26] Susan Folkman and Judith Tedlie Moskowitz. 2000. Stress, positive emotion, andcoping.
Current directions in psychological science
9, 4 (2000), 115–118.[27] Thomas Fritz, Elaine M Huang, Gail C Murphy, and Thomas Zimmermann. 2014.Persuasive technology in the real world: a study of long-term use of activitysensing devices for fitness. In
Proceedings of the SIGCHI conference on humanfactors in computing systems . ACM, 487–496.[28] Mads Frost, Afsaneh Doryab, Maria Faurholt-Jepsen, Lars Vedel Kessing, andJakob E Bardram. 2013. Supporting disease insight through data analysis: re-finements of the monarca self-assessment system. In
Proceedings of the 2013ACM international joint conference on Pervasive and ubiquitous computing . ACM,133–142.[29] Shreyans Gandhi, Maryam Shojaei Baghini, and Soumyo Mukherji. 2015. Men-tal stress assessment-a comparison between HRV based and respiration basedtechniques. In . IEEE, 1029–1032.
HI ’21, May 8–13, 2021, Yokohama, Japan Xianghua and Shuhan, et al. [30] Lisa Graham, Anthony Tang, and Carman Neustaedter. 2016. Help me helpyou: Shared reflection for personal data. In
Proceedings of the 19th InternationalConference on Supporting Group Work . 99–109.[31] Kirsi Halttu and Harri Oinas-Kukkonen. 2017. Persuading to reflect: role ofreflection and insight in persuasive systems design for physical health.
Human–Computer Interaction
32, 5-6 (2017), 381–412.[32] Constance Hammen. 2005. Stress and depression.
Annu. Rev. Clin. Psychol.
Proceedings of the2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing .ACM, 617–621.[34] Javier Hernandez, Pablo Paredes, Asta Roseway, and Mary Czerwinski. 2014.Under pressure: sensing stress of computer users. In
Proceedings of the SIGCHIconference on Human factors in computing systems . ACM, 51–60.[35] Karen Hovsepian, Mustafa al’Absi, Emre Ertin, Thomas Kamarck, Motohiro Naka-jima, and Santosh Kumar. 2015. cStress: towards a gold standard for continuousstress assessment in the mobile environment. In
Proceedings of the 2015 ACMinternational joint conference on pervasive and ubiquitous computing
Human–Computer Interaction
33, 1(2018), 1–48.[44] Matthew Kay. 2014. Challenges in personal health tracking: the data isn’t enough.
XRDS: Crossroads, The ACM Magazine for Students
21, 2 (2014), 32–37.[45] Christina Kelley, Bongshin Lee, and Lauren Wilcox. 2017. Self-tracking formental wellness: understanding expert perspectives and student experiences. In
Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems .ACM, 629–641.[46] Hye-Geum Kim, Eun-Jin Cheon, Dai-Seg Bai, Young Hwan Lee, and Bon-HoonKoo. 2018. Stress and heart rate variability: A meta-analysis and review of theliterature.
Psychiatry investigation
15, 3 (2018), 235.[47] Wendy EJ Knight and Nikki S Rickard. 2001. Relaxing music prevents stress-induced increases in subjective anxiety, systolic blood pressure, and heart rate inhealthy males and females.
Journal of music therapy
38, 4 (2001), 254–272.[48] Kari Kuutti and Liam J Bannon. 2014. The turn to practice in HCI: towards aresearch agenda. In
Proceedings of the SIGCHI Conference on Human Factors inComputing Systems . 3543–3552.[49] Jean Lave, Etienne Wenger, et al. 1991.
Situated learning: Legitimate peripheralparticipation . Cambridge university press.[50] Amanda Lazar, Christian Koehler, Joshua Tanenbaum, and David H Nguyen.2015. Why we use and abandon smart devices. In
Proceedings of the 2015 ACMInternational Joint Conference on Pervasive and Ubiquitous Computing . ACM,635–646.[51] Richard S Lazarus, Anita DeLongis, Susan Folkman, and Rand Gruen. 1985. Stressand adaptational outcomes: The problem of confounded measures. (1985).[52] Richard S Lazarus and Richard S Lazarus. 1991.
Emotion and adaptation . OxfordUniversity Press on Demand.[53] Mark Le Fevre, Jonathan Matheny, and Gregory S Kolt. 2003. Eustress, distress,and interpretation in occupational stress.
Journal of managerial psychology
18, 7(2003), 726–744.[54] Kwangyoung Lee and Hwajung Hong. 2018. MindNavigator: Exploring theStress and Self-Interventions for Mental Wellness. In
Proceedings of the 2018 CHIConference on Human Factors in Computing Systems . ACM, 572.[55] Fenghua Li, Peida Xu, Shichun Zheng, Wenfeng Chen, Yang Yan, Suo Lu, andZhengkui Liu. 2018. Photoplethysmography based psychological stress detectionwith pulse rate variability feature differences and elastic net.
International Journalof Distributed Sensor Networks
14, 9 (2018), 1550147718803298.[56] Ian Li. 2009. Beyond counting steps: Using context to improve monitoring ofphysical activity. (2009).[57] Ian Li, Anind Dey, and Jodi Forlizzi. 2010. A stage-based model of personalinformatics systems. In
Proceedings of the SIGCHI conference on human factors incomputing systems . ACM, 557–566.[58] Ian Li, Anind K Dey, and Jodi Forlizzi. 2012. Using context to reveal factorsthat affect physical activity.
ACM Transactions on Computer-Human Interaction(TOCHI)
19, 1 (2012), 1–21.[59] James J Lin, Lena Mamykina, Silvia Lindtner, Gregory Delajoux, and Henry BStrub. 2006. Fish’n’Steps: Encouraging physical activity with an interactive computer game. In
International conference on ubiquitous computing . Springer,261–278.[60] Wan-Hua Lin, Dan Wu, Chunyue Li, Heye Zhang, and Yuan-Ting Zhang. 2014.Comparison of heart rate variability from PPG with that from ECG. In
TheInternational Conference on Health Informatics . Springer, 213–215.[61] Hong Lu, Denise Frauendorfer, Mashfiqui Rabbi, Marianne Schmid Mast, Gokul TChittaranjan, Andrew T Campbell, Daniel Gatica-Perez, and Tanzeem Choudhury.2012. Stresssense: Detecting stress in unconstrained acoustic environments usingsmartphones. In
Proceedings of the 2012 ACM Conference on Ubiquitous Computing .ACM, 351–360.[62] Yongqiang Lyu, Xiaomin Luo, Jun Zhou, Chun Yu, Congcong Miao, Tong Wang,Yuanchun Shi, and Ken-ichi Kameyama. 2015. Measuring photoplethysmogram-based stress-induced vascular response index to assess cognitive load and stress.In
Proceedings of the 33rd Annual ACM Conference on Human Factors in ComputingSystems . ACM, 857–866.[63] Diana MacLean, Asta Roseway, and Mary Czerwinski. 2013. MoodWings: awearable biofeedback device for real-time stress intervention. In
Proceedingsof the 6th international conference on PErvasive Technologies Related to AssistiveEnvironments . ACM, 66.[64] Marek Malik. 1996. Heart rate variability: Standards of measurement, physio-logical interpretation, and clinical use: Task force of the European Society ofCardiology and the North American Society for Pacing and Electrophysiology.
Annals of Noninvasive Electrocardiology
1, 2 (1996), 151–181.[65] Lena Mamykina, Elizabeth D Mynatt, and David R Kaufman. 2006. Investigatinghealth management practices of individuals with diabetes. In
Proceedings of theSIGCHI conference on Human Factors in computing systems . ACM, 927–936.[66] Karen A Matthews and Brooks B Gump. 2002. Chronic work stress and maritaldissolution increase risk of posttrial mortality in men from the Multiple RiskFactor Intervention Trial.
Archives of Internal Medicine
Proceedings of the 17thInternational Conference on Human-Computer Interaction with Mobile Devices andServices . 86–97.[68] Paolo Melillo, Marcello Bracale, and Leandro Pecchia. 2011. Nonlinear HeartRate Variability features for real-life stress detection. Case study: students understress due to university examination.
Biomedical engineering online
10, 1 (2011),96.[69] Nermine Munla, Mohamad Khalil, Ahmad Shahin, and Azzam Mourad. 2015.Driver stress level detection using HRV analysis. In . IEEE, 61–64.[70] Pablo Paredes and Matthew Chan. 2011. CalmMeNow: exploratory research anddesign of stress mitigating mobile interventions. In
CHI’11 Extended Abstracts onHuman Factors in Computing Systems . ACM, 1699–1704.[71] Misha Patel and Aisling Ann O’Kane. 2015. Contextual influences on the useand non-use of digital technology while exercising at the gym. In
Proceedingsof the 33rd Annual ACM Conference on Human Factors in Computing Systems .2923–2932.[72] Thomas G. Pickering. 2001. Mental stress as a causal factor in the developmentof hypertension and cardiovascular disease.
Current Hypertension Reports
3, 3(2001), 249–54.[73] Bernd Ploderer, Wolfgang Reitberger, Harri Oinas-Kukkonen, and Julia vanGemert-Pijnen. 2014. Social interaction and reflection for behaviour change.[74] Sharon Praissman. 2008. Mindfulness-based stress reduction: A literature reviewand clinician’s guide.
Journal of the American Academy of Nurse Practitioners
Extended Abstracts of the 2018 CHI Conferenceon Human Factors in Computing Systems . ACM, LBW599.[76] Amon Rapp and Federica Cena. 2016. Personal informatics for everyday life:How users without prior self-tracking experience engage with personal data.
International Journal of Human-Computer Studies
94 (2016), 1–17.[77] Amon Rapp, Alessandro Marcengo, Luca Buriano, Giancarlo Ruffo, Mirko Lai, andFederica Cena. 2018. Designing a personal informatics system for users withoutexperience in self-tracking: a case study.
Behaviour & Information Technology
Proceedings of the 2017 CHI Conferenceon Human Factors in Computing Systems . ACM, 6864–6875.[79] John Rooksby, Mattias Rost, Alistair Morrison, and Matthew Chalmers. 2014.Personal tracking as lived informatics. In
Proceedings of the SIGCHI conference onhuman factors in computing systems . ACM, 1163–1172.[80] Annika Rosengren, Steven Hawken, Stephanie Ôunpuu, Karen Sliwa, MohammadZubaid, Wael A Almahmeed, Kathleen Ngu Blackett, Chitr Sitthi-amorn, HiroshiSato, Salim Yusuf, et al. 2004. Association of psychosocial risk factors with ata Engagement Reconsidered: A Study of Automatic Stress Tracking Technology in Use CHI ’21, May 8–13, 2021, Yokohama, Japan risk of acute myocardial infarction in 11119 cases and 13648 controls from 52countries (the INTERHEART study): case-control study.
The Lancet
Proceedings of the 8th ACM conference on designing interactive systems . ACM,47–56.[82] Hans Selye. 1965. The stress syndrome.
The American Journal of Nursing (1965),97–99.[83] Moushumi Sharmin, Andrew Raij, David Epstien, Inbal Nahum-Shani, J GayleBeck, Sudip Vhaduri, Kenzie Preston, and Santosh Kumar. 2015. Visualizationof time-series sensor data to inform the design of just-in-time adaptive stressinterventions. In
Proceedings of the 2015 ACM International Joint Conference onPervasive and Ubiquitous Computing . ACM, 505–516.[84] Petr Slovák, Christopher Frauenberger, and Geraldine Fitzpatrick. 2017. Reflectivepracticum: A framework of sensitising concepts to design for transformative re-flection. In
Proceedings of the 2017 CHI Conference on Human Factors in ComputingSystems . 2696–2707.[85] Andrew Steptoe. 1991. Invited review: The links between stress and illness.
Journal of psychosomatic research
35, 6 (1991), 633–644. [86] Feng-Tso Sun, Cynthia Kuo, Heng-Tze Cheng, Senaka Buthpitiya, Patricia Collins,and Martin Griss. 2010. Activity-aware mental stress detection using physiolog-ical sensors. In
International conference on Mobile computing, applications, andservices . Springer, 282–301.[87] Tencent. 2018. 2015 Smart Wearable Devices Market White Paper. https://cloud.tencent.com/developer/article/1154297[88] Jonathan Tudge. 1992. Vygotsky, the zone of proximal development, and peercollaboration: Implications for classroom practice. (1992).[89] Rubén Vicente-Sáez and Clara Martínez-Fuentes. 2018. Open Science now: Asystematic literature review for an integrated definition.
Journal of businessresearch
88 (2018), 428–436.[90] Elena Vildjiounaite, Johanna Kallio, Vesa Kyllönen, Mikko Nieminen, IlmariMäättänen, Mikko Lindholm, Jani Mäntyjärvi, and Georgy Gimel’farb. 2018.Unobtrusive stress detection on the basis of smartphone usage data.
Personaland Ubiquitous Computing
22, 4 (2018), 671–688.[91] Lisa M Vizer. 2009. Detecting cognitive and physical stress through typingbehavior. In
CHI’09 Extended Abstracts on Human Factors in Computing Systems .ACM, 3113–3116.[92] Bin Yu, Mathias Funk, Jun Hu, and Loe Feijs. 2017. StressTree: a metaphoricalvisualization for biofeedback-assisted stress management. In