Measuring Self-Esteem with Passive Sensing - Koustuv Saha (PDF)

Measuring Self-Esteem with Passive Sensing Mehrab Bin Morshed*, Koustuv Saha*, Munmun De Choudhury, Gregory D. Abowd, Thomas Plötz {mehrab.morshed,koustuv.saha,munmund,abowd,thomas.ploetz}@gatech.edu Georgia Institute of Technology Atlanta, Georgia, U.S. ABSTRACT Self-esteem encompasses how individuals evaluate themselves and is an important contributor to their success. Self-esteem has been traditionally measured using survey-based methodologies. How- ever, surveys suffer from limitations such as retrospective recall and reporting biases, leading to a need for proactive measurement approaches. Our work uses smartphone sensors to predict self- esteem and is situated in a multimodal sensing study on college students for five weeks. We use theory-driven features, such as phone communications and physical activity to predict three di- mensions, performance, social, and appearance self-esteem. We conduct statistical modeling including linear, ensemble, and neural network regression to measure self-esteem. Our best model pre- dicts self-esteem with a high correlation (𝑟) of 0.60 and low SMAPE of 7.26% indicating high predictive accuracy. We inspect the top features finding theoretical alignment; for example, social interac- tion significantly contributes to performance and appearance-based self-esteem, whereas, and physical activity is the most significant contributor towards social self-esteem. Our work reveals the effi- cacy of passive sensors for predicting self-esteem, and we situate our observations with literature and discuss the implications of our work for tailored interventions and improving wellbeing. CCS CONCEPTS • Human-centered computing → Emp irical studies in ubiquitous and mobile comp uting; • Applied computing → Psychology. KEYWORDS self-esteem, passive sensing, college students, wellbeing, campuslife ACM Reference Format: Mehrab Bin Morshed*, Koustuv Saha*, Munmun De Choudhury, Gregory D. Abowd, Thomas Plötz. 2020. Measuring Self-Esteem with Passive Sensing. In 14th EAI International Conference on Pervasive Comp uting Technologies for Healthcare (PervasiveHealth ’20), May 18–20, 2020, Atlanta, GA, USA. ACM, New York, NY, USA, 4 pages. https://doi.org/10.1145/3421937.3421952 1 INTRODUCTION AND BACKGROUND A musician must make music, an artist must paint, a poet must write, if he is to be ultimately at peace with himself. What a man can be, he must be. This need we may call self-actualization —Maslow 1981 *Both authors contributed equally to this work. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from [email protected]. PervasiveHealth ’20, May 18–20, 2020, Atlanta, GA, USA © 2020 Association for Computing Machinery. ACM ISBN 978-1-4503-7532-0/20/05...$15.00 https://doi.org/10.1145/3421937.3421952 Psychologist Abraham Harold Maslow in his seminal book, Moti- vation and Personality, identified various human needs that range from survival to intellectual growth [22]. Such needs, often re- ferred to as Maslow’s hierarchy of needs can be clustered into five categories—Physiological, Safety, Love or Belonging, Self-Esteem, and Self-Actualization (Figure 1). Figure 1: Maslow’s Hierarchy of Needs as per Maslow [21] Self-esteem is considered to be a form of a con- cept that encom- passes how an in- dividual evaluates themselves on a scale ranging from positive (or self- affirming) to neg- ative (or self- denigrating) [18]. In other words, self-esteem corre- sponds to some- one evaluating themselves in the form of how much they like or dislike themselves [3]. This process of self-evaluation continues throughout the lifetime of an individual, and it is aided by a variety of social, cultural, and environmental constraints. Self- esteem is one of the basic human needs, and it is known to sig- nificantly contribute to one’s motivation and success [21]. Low self-esteem can affect performance at school or work, and high self-esteem can help an individual navigate life by positive attitude and belief in achieving their goals. Individuals with damaged and lower self-esteem are at a greater risk of psychosocial distress, and maybe vulnerable to the demanding circumstances of day-to-day life [19]. An early assessment of one’s self-esteem can not only facil- itate one to reflect on their capabilities, but also can guide tailored interventions towards uplifting their self-esteem, motivation, desire to perform, and in turn their wellbeing. Understanding self-esteem can foster adopting preemptive steps to facilitate the psychological and cognitive needs of individuals. Over the years, researchers from the domain of psychology, sociology, cognitive sciences, etc., have come up with different methodologies to estimate one’s self-esteem [3]. Heatherton and Polivy proposed self-esteem to consist of three primary dimensions: performance, social, and physical appearance [17]. Rosenberg’s scale is a popular way to measure self-esteem among a variety of targets, and it measures the self-esteem of individuals in abso- lute terms [28]. Miyamoto and Dornbush proposed an alternate self-esteem measurement scale that focuses on more individual as- pects (e.g., intelligence, physical attractiveness, etc.) [24], and Ziller  PervasiveHealth ’20, May 18–20, 2020, Atlanta, GA, USA Bin Morshed and Saha, et al. Table 1: Descriptive statistics of the DASS-21 data. Levels inferred as per Gomez [14]. Level Dep. Anx. Str. Normal 26 27 26 Mild 5 6 9 Moderate 8 10 8 Severe 3 0 1 Extremely Severe 3 2 1 Table 2: Descriptive statistics of the EMA data. Metric Value # Participants 51 # Responses 1,606 Mean Responses 31.49 Median Responses 28.00 StDev. Responses 21.13 et al. proposed a social self-esteem scale where individuals evaluate themselves in their social circle [48]. Traditionally, interviews and surveys have been the primary instruments to quantify psychological constructs [9]. These ap- proaches are highly reliant on a respondent’s retrospective recall and subjective assessments [13]. Aggregating multiple events over the past leads to difficulty in recollection, leading to poor quality of recorded data. Some of the confounds of static self-reported data can be mitigated by using in-situ data collection approaches [34]. One popular approach is using ecological momentary assessments (EMAs), also known as experience sampling [10, 35]. EMAs have many advantages over traditional research designs for character- izing complex psychological processes [39, 40], and has been a promising approach in longitudinal studies facilitating actively sensed behaviors and moods [36, 43]. This approach works by prompting participants to respond to survey items in-the-moment and within their natural context. With the advances in mobile active-sensing (e.g, through smart- phones), EMAs can now be conducted at scale. Accordingly, EMAs are now extensively used in large-scale multimodal sensing stud- ies in ubiquitous computing research [23, 27, 45]. Within the Stu- dentLife Project, Wang et al. used mobile EMAs to capture ground- truth information about student activity, emotions, contexts, etc [45]. Although better than static survey instruments on many fronts, active sensing comes with limitations of scale, access, and affor- dance [35]. EMAs often disseminated through prompts induce a response burden on participants through disruptions [41]. This leads to a tradeoff between balancing the construct validity of par- ticipant responses and their compliance [7]. Also, towards a more proactive and holistic understanding of individual wellbeing, re- searchers have recently valued passive sensing modalities [4, 45]. The unobtrusive and low burden nature of passively sensed data complement actively sensed data. With the ubiquity and widespread use of smartphones and wearables, passive sensing modalities en- able a cheap and easier mechanism to capture longitudinal and dense human behavior at scale [44–46]. This paper asks, Can we automatically and scalably p redict self- esteem using p assive sensing modalities available on commodity de- vices? Our study conducts an EMA-based survey of self-esteem drawing on Heatherton’s State Self-Esteem Scale [17]. We build statistical models to predict self-esteem using passively sensed data such as physical activity, conversation patterns, and digital com- munication patterns. We conduct a deeper dive into the predictive features and adopt a theory-driven approach to understand the importance of the significant features in the context of self-esteem. 2 STUDY AND DATA This paper leverages data collected in the CampusLife project at Georgia Tech [7, 31]. The data collection was conducted for five weeks during Spring 2016 (March-April) by enrolling several stu- dents at Georgia Tech, a large public university in the U.S., and the study was approved by the Institutional Review Board at Georgia Tech. This study enrolled 51 students, consisting of 40% females and 60% males, and 46% undergraduate and 54% graduate students. The mean age of participants was 22 years. The CampusLife study provided the participants with smart- phones that were instrumented to collect a variety of actively and passively sensed data. For active sensing, the smartphones used the Quedget platform that sent ecological momentary assessments in the form of daily state-based survey constructs including affect states, self-esteem, stress, etc. For passive sensing, smartphone- based sensor data was collected in the form of physical activity, frequency of incoming and outgoing calls, frequency of messages, conversation inferences of individuals, WiFi access points, etc. In addition, the participants also answered one-time surveys on their individual differences (e.g., demographics) and wellbeing-trait based measures (Perceived Stress Scale (PSS) [8], Flourishing Scale, and Depression Anxiety and Stress Scale (DASS-21) [1]). For the ease of exposition, we provide a brief summary of the student mental wellbeing in terms of DASS-21 based mental health assessment in Table 1. Given that roughly half the study population shows above-normal levels of mental health, this dataset can be assumed to contain a variety of representation in terms of mental well-being. For self-esteem, we adopted the Heatherton’s State Self-Esteem Scale as a Self-Esteem EMA employed on Quedget. These EMAs asked the participants to reflect on their academic performance (scholastic ability), social (e.g., feeling good, feeling self-conscious), and physical appearance based self-esteem. 3 METHODS AND RESULTS 3.1 Measuring Self-Esteem from EMAs First, we compile our groundtruth dataset of measuring self-esteem from ecological momentary assessment (EMA) responses from par- ticipants. The EMAs correspond to daily state-based self-esteem measurement, along three dimensions — 1) Performance, 2) So- cial, and 3) Appearance [17]. Based on these three dimensions, we measure daily self-esteem per individual as a state-measure. Addi- tionally, we aggregate the self-esteem assessments per individual throughout the duration of the study. We draw upon the literature to evaluate the convergent validity of our self-esteem measurement. Prior work reports self-esteem to show a correlation with mood in a range of 0.40 and 0.60 [6, 17], and the average correlation in our dataset is 0.52 with statistical sig- nificance (see Table 3). This provides reliability in our groundtruth assessment of self-esteem. Figure 2 shows a distribution of state self-esteem within our dataset, we find that it averages at 2.41 (std. = 0.97) for performance, 2.84 (std. = 0.55) for social, 2.17 (std. = 1.07) for appearance per participant. 3.2 Measuring Self-Esteem with Passively Sensed Data We aim at scalably inferring the self-esteem using passive sensing data. Adopting a theory-driven approach, we build machine learn- ing classifiers of self-esteem. As any other classification methodol- ogy, our approach involves tuning both machine learning algorithm (and corresponding parameters) and the sensing features.  Measuring Self-Esteem with Passive Sensing PervasiveHealth ’20, May 18–20, 2020, Atlanta, GA, USA Table 3: Pearson’s𝑟 betwen EMA-based self-esteem and DASS-21 scores. DASS Performance Social Appearance Depression 0.60*** 0.51*** 0.57*** Anxiety 0.38** 0.51*** 0.65*** Stress 0.40* 0.40*** 0.60*** 1 2 3 4 5 Mean Self-Esteem 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Density (Users) Performance Social Appearance Figure 2: Distribu- tion of self-esteem in EMA data. 3.2.1 Passive Sensing Features. This paper uses a variety of pas- sively sensed features captured by sensors on smartphones. We adopt a theory-driven approach to consider a feature space that is situated in prior literature on psychological contructs, wellbeing, and self-esteem [3]. We describe our features below. Physical Activity. We consider physical activity based features drawing on two streams of prior research — 1) theoretically, self- esteem is known to be associated with physical movement and activity [33, 42], and 2) ubiquitous computing research has revealed the potential of passively sensed physical activity to understand wellbeing and human dynamics [11, 12, 25]. We use state-of-the- art google activity recognition API that records activity in various states such as walking, biking, sitting, etc [2]. Conversational Setting. Because social and environmental con- text of an individual bears a potential link to one’s self-esteem [37], we use microphone-captured conversation signals as one of our features. This datastream infers if at a given point of time, an indi- vidual is situated in a social setting. Our approach does not include storing or using any sensitive voice data, rather the inference of social conversation events. Communication Pattern. Because social and environmental con- text of an individual bears a potential link to one’s self-esteem [37], we use communication signals as one of our features. Specifically, we use the total number of calls and frequency of messages per day. 3.2.2 Prediction Model. To predict self-esteem from passively sensed features scalably, we adopt regression models. We use individual- wise mean, median, and standard deviation per sensing modality as independent variables to predict corresponding self-esteem mea- sures. We build three separate models for three self-esteem com- ponents, academic performance, social, and appearance. We try a variety of algorithms, particularly, linear regression (LR), gradi- ent boosted regression (GBR), and deep learning based multilayer perceptron (MLP). These algorithm choices are motivated by the no- tion that these cover a broad spectrum of algorithm families spread across linear and non-linear regression, decision trees and ensemble learning, and deep neural networks. We tune model parameters and evaluate our models using a leave-one-out cross-validation ap- proach and measure goodness of fit as 𝑅2. To evaluate our models, we consider a pooled correlation (𝑟) and pooled symmetric mean absolute percentage error (SMAPE), which is computed as the mean percentage relative difference between predicted and actual values over a mean of the two values [20, 29]. SMAPE values range at 0-100%, and lower values of error indicate better predictive ability. 3.2.3 Prediction Performance. Table 4 reports the performance met- rics of the above models. We find that the GBR model outperforms Table 4: Pooled performance metrics as leave-one-out cross- validation. (***𝑝<0.0001, **p<0.001) Performance Social Appearance Model R2 𝑟 SMAPE R2 𝑟 SMAPE R2 𝑟 SMAPE LR 0.29 0.36*** 18.57 0.61 0.54*** 14.32 0.49 0.38*** 15.32 GBR 0.46 0.42** 8.61 0.83 0.77*** 5.64 0.79 0.59*** 7.53 MLP 0.35 0.41*** 13.62 0.68*** 0.63*** 10.23*** 0.62 0.43** 10.37 G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G 1 2 3 4 5 1 2 3 4 5 Actual Performance SE Predicted Performance SE (a) Performance G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G 1 2 3 4 5 1 2 3 4 5 Actual Social SE Predicted Social SE (b) Social G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G G 1 2 3 4 5 1 2 3 4 5 Actual Appearance SE Predicted Appearance SE (c) Appearance Figure 3: Scatter plots of actual and predicted values of the three kinds of self-esteem. Table 5: Relative feature importance in the GBR model. F eatures Performance Social Appearance Calls 0.29*** 0.28*** 0.26*** Texts 0.27*** 0.11*** 0.31*** Conversations 0.21*** 0.25*** 0.15*** Phy. Activity 0.23*** 0.37*** 0.28*** all both in terms of low SMAPE and high correlation with statistical significance. GBR performs at an SMAPE=8.61% and 𝑟=0.42 for per- formance, SMAPE=5.64% and 𝑟=0.77 for social, and SMAPE=7.53% and 𝑟=0.59 for appearance based self-esteem (ref: Figure 3). 3.2.4 Feature-Model Relevance Interpretation. For the best perform- ing (GBR) model, we use K-best univariate statistical scoring model using mutual information to obtain the relative importance among features and establish their statistical significance using ANOVA to obtain the top features, which are reported in Table 5. We find that many digital behavioral categories are significant, which also aligns with prior literature on a self-esteem [38, 47]. For example, calls, texts, and conversations are proxies for so- cial interaction. For predicting social self-esteem, we can see that features from calls, conversations, and physical activities are the strongest predictors. Several studies suggest that low social self- esteem often leads to isolation and a lack of communication [5]. In the case of predicting appearance, we can see that calls, messages, and physical activities are most predictive. Hayes et al. found that self-esteem related to physical appearance was strongly correlated with physical activities undertaken by the individual [16]. 4 DISCUSSION AND CONCLUSION This paper adopted a statistical modeling-based approach to pre- dict three kinds of self-esteem (academic, social, and appearance) to find that passive sensing modalities can reliably predict self- esteem. These approaches that rely on passively sensed data can help not only address challenges associated with traditional modes of data collection but also help us make assessments with a contin- ual and longitudinal fashion. When enacted on a situated commu- nity such as college campuses and student populations, these ap- proaches can help student-wellbeing-stakeholders understand how  PervasiveHealth ’20, May 18–20, 2020, Atlanta, GA, USA Bin Morshed and Saha, et al. student psychological constructs change (both at the individual- and community-level) over time in both normalcy and crisis [15]. Limitations and F uture Directions Our work has limitations, many of which open up opportunities for future research. Our study concerns college students, and our results may not generalize to other populations, and we refrain from making generalized claims within and beyond college campuses because of self-selection and population biases. In one way, we conduct a feasibility study that reveals the potential of passive sensing in predicting self-esteem. Future work can conduct similar studies with a larger participation pool and over a longer duration of time, taking advantage of more newly available commercial smart devices and sensing modalities, e.g., health tracker, smartwatches [23, 30]. While we considered physical passive sensors, prior work has also shown the potential of unobtrusively obtained social media and online digital footprints in measuring self-esteem [26, 47] and psychological constructs in situated communities like college campuses [32]. It would be inter- esting to examine how all of these modalities may complementarily function together towards measuring psychological constructs. 5 ACKNOWLEDGMENTS Bin Morshed was partly supported by a grant from Siemens Fu- tureMaker Fellowship Task Order #7. De Choudhury was partly supported by NIH grant #R01MH117172. The project was partly supported by a grant from Semiconductor Research Corporation in collaboration with Intel Labs. REFERENCES [1] Martin M Antony, Peter J Bieling, Brian J Cox, Murray W Enns, and Richard P Swinson. 1998. Psychometric properties of the 42-item and 21-item versions of the Depression Anxiety Stress Scales in clinical groups and a community sample. Psychological assessment 10, 2 (1998), 176. [2] Activity Recognition API. 2018. https://developers.google.com/location-context/ activity-recognition/. Accessed: 2018-11-01. [3] Robert W Baker and Bohdan Siryk. 1984. Measuring adjustment to college. Journal of counseling p sychology 31, 2 (1984), 179. [4] Mehrab Bin Morshed, Samruddhi S Kulkarni, Richard Li, Koustuv Saha, Lama Nachman, Hong Lu, Leah Galante Roper, Lucia Mirabella, Sanjeev Srivastava, Munmun De Choudhury, Kaya de Barbaro, Thomas Plötz, and Gregory D. Abowd. 2020. A Real-Time Eating Detection System for Capturing Eating Moments and Triggering EMAs to Collect Further Context. (2020). [5] Forough Bozorgpour and Azimeh Salimi. 2012. State self-esteem, loneliness and life satisfaction. Procedia-Social and Behavioral Sciences 69 (2012), 2004–2008. [6] Joel Brockner. 1983. Low self-esteem and behavioral plasticity: Some implications. Review of p ersonality and social p sychology 4 (1983), 237–271. [7] Larry Chan, Vedant Das Swain, Christina Kelley, Kaya de Barbaro, Gregory D Abowd, and Lauren Wilcox. 2018. Students’ Experiences with Ecological Mo- mentary Assessment Tools to Report on Emotional Well-being. IMWUT (2018). [8] Sheldon Cohen, T Kamarck, R Mermelstein, et al. 1994. Perceived stress scale. Measuring stress: A guide for health and social scientists (1994). [9] Rex William Cowdry, David L Gardner, et al. 1991. Mood variability: A study of four groups. The American Journal of Psychiatry 148, 11 (1991), 1505. [10] Mihaly Csikszentmihalyi and Reed Larson. 2014. Validity and reliability of the experience-sampling method. In Flow and the foundations of p ositive p sychology. [11] Vedant Das Swain et al. 2019. A Multisensor Person-Centered Approach to Understand the Role of Daily Activities in Job Performance with Organizational Personas. Proc. ACM IMWUT (2019). [12] Vedant Das Swain, Hyeokhyen Kwon, Bahador Saket, Mehrab Bin Morshed, Kathy Tran, Devashru Patel, Yexin Tian, Joshua Philipose, et al. 2020. Leveraging WiFi Network Logs to Infer Social Interactions: A Case Study of Academic Performance and Student Behavior. arXiv p rep rint arXiv:2005.11228 (2020). [13] Barbara L Fredrickson. 2000. Extracting meaning from past affective experiences: The importance of peaks, ends, and specific emotions. Cogn. Emot. (2000). [14] Fernando Gomez. 2016. A Guide to the Depression, Anxiety and Stress Scale (DASS 21). [15] Penny Haney and Joseph A Durlak. 1998. Changing self-esteem in children and adolescents: A meta-analytical review. Journal of clinical child p sychology (1998). [16] Sean D Hayes, Peter RE Crocker, and Kent C Kowalski. 1999. Gender differences in physical self-perceptions, global self-esteem and physical activity: Evaluation of the physical self-perception profile model. Journal of Sp ort Behavior (1999). [17] Todd F Heatherton and Janet Polivy. 1991. Development and validation of a scale for measuring state self-esteem. J. Pers. Soc. Psychol (1991). [18] John P Hewitt. 2002. The social construction of self-esteem. Handbook of p ositive p sychology s 135147 (2002). [19] Mizuho Hosogi, Ayumi Okada, Chikako Fujii, Keizou Noguchi, and Kumi Watan- abe. 2012. Importance and usefulness of evaluating self-esteem in children. BioPsychoSocial medicine (2012). [20] Vladik Kreinovich, Hung T Nguyen, and Rujira Ouncharoen. 2014. How to estimate forecasting quality: a system-motivated derivation of symmetric mean absolute percentage error (SMAPE) and other similar characteristics. (2014). [21] Abraham Harold Maslow. 1943. A theory of human motivation. Psychol. Rev (1943). [22] Abraham Harold Maslow. 1981. Motivation and p ersonality. Prabhat Prakashan. [23] Stephen M. Mattingly et al. 2019. The Tesserae Project: Large-Scale, Longitudinal, In Situ, Multimodal Sensing of Information Workers. In CHI Ext. Abstracts. [24] S Frank Miyamoto and Sanford M Dornbush. 1956. A test of interactionist hypotheses of self-conception. Amer. J. Sociology 61, 5 (1956), 399–403. [25] Mehrab Bin Morshed, Koustuv Saha, Richard Li, Sidney K. D’Mello, Munmun De Choudhury, Gregory D. Abowd, and Thomas Plötz. 2019. Prediction of Mood Instability with Passive Sensing. PACM IMWUT (2019). [26] Carly Renee Pineiro. 2016. Social media use and self-esteem in undergraduate students. (2016). [27] Mashfiqui Rabbi, Min Hane Aung, Mi Zhang, and Tanzeem Choudhury. 2015. MyBehavior: automatic personalized health feedback from user behaviors and preferences using smartphones. In Proc. Ubicomp . [28] Morris Rosenberg. 1965. Rosenberg self-esteem scale (RSE). Accep tance and commitment therap y. Measures p ackage 61, 52 (1965), 18. [29] Koustuv Saha et al. 2019. Imputing Missing Social Media Data Stream in Multi- sensor Studies of Human Behavior. In Proc. ACII. [30] Koustuv Saha et al. 2019. Social Media as a Passive Sensor in Longitudinal Studies of Human Behavior and Wellbeing. In CHI Ext. Abstracts. ACM. [31] Koustuv Saha, Larry Chan, Kaya De Barbaro, Gregory D Abowd, and Munmun De Choudhury. 2017. Inferring mood instability on social media by leveraging ecological momentary assessments. PACM IMWUT (2017). [32] Koustuv Saha and Munmun De Choudhury. 2017. Modeling stress with social media around incidents of gun violence on college campuses. PACM HCI (2017). [33] Rob Sands, Julie Tricker, Cheyne Sherman, Christine Armatas, and Wayne Maschette. 1997. Disordered eating patterns, body image, self-esteem, and physi- cal activity in preadolescent school children. Int. J. Eat. Disord. (1997). [34] Norbert Schwarz. 2007. Retrospective and concurrent self-reports: The rationale for real-time data capture. The science of real-time data cap ture: Self-rep orts in health research (2007). [35] Christie Napa Scollon, Chu-Kim Prieto, and Ed Diener. 2009. Experience sampling: promises and pitfalls, strength and weaknesses. In Assessing well-being. [36] Saul Shiffman, Arthur A Stone, and Michael R Hufford. 2008. Ecological momen- tary assessment. Annu. Rev. Clin. Psychol. 4 (2008), 1–32. [37] Roberta G Simmons, Dale A Blyth, Edward F Van Cleave, and Diane Mitsch Bush. 1979. Entry into early adolescence: The impact of school structure, puberty, and early dating on self-esteem. American Sociological Review (1979), 948–967. [38] Charles Steinfield, Nicole B Ellison, and Cliff Lampe. 2008. Social capital, self- esteem, and use of online social network sites: A longitudinal analysis. (2008). [39] Arthur A Stone, Joseph E Schwartz, John M Neale, Saul Shiffman, Christine A Marco, Mary Hickcox, Jean Paty, Laura S Porter, and Laura J Cruise. 1998. A comparison of coping assessed by ecological momentary assessment and retro- spective recall. Journal of p ersonality and social p sychology 74, 6 (1998), 1670. [40] Arthur A Stone and Saul Shiffman. 1994. Ecological momentary assessment (EMA) in behavorial medicine. Annals of Behavioral Medicine (1994). [41] Hyewon Suh, Nina Shahriaree, Eric B Hekler, and Julie A Kientz. 2016. Developing and Validating the User Burden Scale: A Tool for Assessing User Burden in Computing Systems. In Proc. CHI. [42] Mark S Tremblay, J Wyatt Inman, and J Douglas Willms. 2000. The relationship between physical activity, self-esteem, and academic achievement in 12-year-old children. Pediatric exercise science 12, 3 (2000), 312–323. [43] Timothy J Trull and Ulrich W Ebner-Priemer. 2009. Using experience sampling methods/ecological momentary assessment (ESM/EMA) in clinical assessment and clinical research: introduction to the special section. (2009). [44] N Van Saane, Judith K Sluiter, JHAM Verbeek, and MHW Frings-Dresen. 2003. Reliability and validity of instruments measuring job satisfaction—a systematic review. Occup ational medicine 53, 3 (2003), 191–200. [45] Rui Wang et al. 2014. StudentLife: assessing mental health, academic performance and behavioral trends of college students using smartphones. In Ubicomp . [46] Weichen Wang, Gabriella M Harari, Rui Wang, Sandrine R Müller, Shayan Mirja- fari, Kizito Masaba, and Andrew T Campbell. 2018. Sensing Behavioral Change over Time: Using Within-Person Variability Features from Mobile Sensing to Predict Personality Traits. IMWUT (2018). [47] Anis Zaman, Rupam Acharyya, Henry Kautz, and Vincent Silenzio. 2019. Detect- ing low self-esteem in youths from web search data. In WebConf. [48] Robert C Ziller, Joan Hagey, Mary Smith, and Barbara H Long. 1969. Self-esteem: a self-social construct. J. Consult. Clin. Psychol (1969).