The relationship between workplace stress and computer use has mostly been investigated with self-reports or in controlled environments. However, self-report methods are prone to memory and emotion expression biases, and can be interruptive to employees when implemented for continuous stress tracking in real workplace environments. Researchers have explored the use of wearable sensors for unobtrusive and continuous stress tracking, but mostly in controlled laboratory settings, which limit the understanding of factors influencing stress in real-workplace environments, and the extent to which passive sensing can reveal information about stress during uncontrolled computer interactions.

This dissertation presents novel findings on computer use and stress at the workplace by employing computational methods leveraging computer activity logging and wearable devices that unobtrusively and continuously measured physiological stress through heart-rate variability in two real-world workplace settings: information work and medical work.

In the first part of the dissertation, fifty office employees were tracked for three to four weeks. Time spent on the work computer during and outside workhours, email work strategy, window switching, and computer activity types explained 14\% of the variance in the daily stress duration. Individual differences (personality and work-life balance) moderate the relationship between workplace computer use factors and stress. A novel measure of variability in daily computer work was associated with perceived job demands, effort and overcommitment and arousal. Employees’ perspectives on technology-supported stress tracking at the workplace indicated trust in algorithmic output, confirmation bias, and challenges balancing unobtrusiveness and engagement.

The second part of the dissertation analyzed how physicians use Electronic Health Record (EHR) systems and measured their physiological stress throughout the workday. One month of EHR logs of 1275 physicians were analyzed to characterize EHR use. Temporal patterns of EHR inbox use were found to be different from other EHR functions in their distribution throughout the day. Factors associated with high EHR inbox use were identified. Physiological stress data were collected for 47 physicians for a week and paired with their EHR logs. Among three patterns of EHR inbox work identified, the pattern characterized by working mostly outside of workhours had the longest average stress duration. Inbox work duration, the rate of EHR window switching, working outside of workhours and batching inbox work were associated with physicians’ daily stress duration.

By evaluating a range of computer use factors and their association with daily physiological stress, the dissertation extends previous work that often focused on specific computer tasks or used self-reports. I provide recommendations and design implications for supporting different personal and organizational technology-supported stress tracking goals, and suggest future areas of work.

Young people today grow up alongside the rapidly developing information and communication technologies, which have transformed their informational habits and have allowed for a culture of autonomous exploration and learning. Widely adopted by college students, Facebook serves multi-faceted purposes in college life. While some research claims that Facebook is detrimental to learning by distracting students (from formal learning), others see great potential in online social networking platforms to support (informal) learning opportunities. This dissertation responds to these contrasting views on the uses and effects of Facebook, particularly with respect to learning. Learning, here, is conceptualized as learners constructing knowledge using various methods and pathways in a social and cultural context.

This dissertation investigated: 1) when and under what conditions college students actually use Facebook, and how; 2) whether and in what ways college students’ informal learning activities on Facebook are consistent with elements of connected learning—personal interest, peer support, and academic orientation; and 3) how college students perceive the benefits and shortcomings associated with news engagement on Facebook. To answer these questions, I studied in situ Facebook use of undergraduate college students using mixed methods, including automatic logging, experience sampling surveys, diaries, and semi-structured interviews.

Students exhibited two contrasting Facebook use patterns after schoolwork and after leisure activities. Results suggest that a student’s attentional state might carry over from their previous activity into a Facebook use session. Focusing on students’ self-directed learning activities, findings show that connected learning can take place in spaces that are not designed for educational purposes, like Facebook. Specifically, the networking environment provides a foundation to support personal interests, which allow for interest-driven academic activities on Facebook, bridging the formal and informal learning divide. We need to caution, however, that learning can also be hindered by 1) students’ lack of critical information skills, 2) information filtering, personalization, and the overall quality of information on Facebook, and 3) the challenges surrounding self-presentation due to context collapse on Facebook. Considering both the opportunities and pitfalls, I provide directions for students and educators on how to effectively incorporate Facebook into meaningful college learning experience.

A mixed methods study was conducted to understand undergraduate students’ victimization, perceptions, and reporting of cybercrimes. Semi-structured interviews were conducted with 10 participants that provided the groundwork for questions that would be asked in an online survey. A total of 222 survey responses were collected from which four linear regression models were built to predict for perceptions (i.e. fear and concern) of cybercrimes, preventative measures that can mitigate cybercrime victimization, enabling behaviors that can lead to cybercrime victimization, and the likelihood of cybercrime reporting. The results of the models were that: (1) media exposure, cybercrime knowledge, and the harmfulness of cybercrimes were associated with cybercrime perceptions; (2) self-control was associated with preventative measures and enabling behaviors; and (3) cybercrime perceptions were associated with enabling behaviors and the likelihood of cybercrime reporting. Participants reported acquiring their knowledge of cybercrimes predominantly through personally knowing someone who had been victimized by a cybercrime (i.e. vicarious cybercrime experiences) and the media. The survey results show that undergraduate students reported being predominantly victimized by malware, hacking, and phishing. Despite the fact that both interviewees and survey participants expressed that they found reporting cybercrimes and having access to cybercrime victimization statistics to be important, the majority of participants did not know how to officially report a cybercrime. Cybercrime victimization can cause various kinds of harm (e.g. psychological, social, financial) to those affected. The focus on how cybercrimes affect a young subject group like undergraduate students who use technology frequently is motivated by the goal to bring more awareness to this issue and for preventative measures to be taken early on to help mitigate cybercrime risk.

Despite heavy investments by local and national governments, crime continues to remain a serious problem in the society. Current state-of-the-art is below average and relies heavily on crime intelligence and efforts of human personnel across several law enforcement agencies.

Over the last few years, crowdsourcing has been promoted worldwide through several Suspicious Activity Reporting (SAR) campaigns. However, such systems have proven to be largely unsuccessful owing to complicated reporting processes, excessive noise, limited authorized personnel, lack of motivation and one-way communication.

To address these issues and improve counterintelligence, this thesis proposes two approaches: (i) Combination of Machine Learning and Natural Language Processing to promote generic reporting while automating the process of crime detection, summarisation and delegation (ii) Improving crowd intelligence by enabling better citizen involvement through newer information sharing techniques, sophisticated sensors and responsive feedback. In this thesis, we also investigate lone wolf terrorism as a special case of crime, and evaluate application of our proposed methods for its mitigation.

Over the last decade, a large subset of cyberpsychology research has investigated the relationship between technology (e.g. smartphone, internet, and social media) use in young adults, and individual differences in a range of psychological constructs such as satisfaction with life (SWL), locus of control, impulsivity, narcissism, loneliness, and depression, anxiety, and stress (DAS). However, this body of work has almost exclusively relied on the use of self-reports when measuring technology use. Self-reports of technology use have been previously shown to have a weak relationship with the objective behaviors that these measures aim to capture. As a result, some have feared that the extant findings in this field may be invalid or unreliable in terms of alignment with objectively measured behaviors. Furthermore, the absence of behavioral sensing methods in this field also suggests that little is known about the precise patterns of technology use that may characterize individual differences in these constructs.

My dissertation is centered around a large behavioral sensing study of 271 young adults, wherein participants had their smartphone and personal computer activity logged continuously for a month long period, and completed a wide set of different psychological surveys, technology use surveys, as well as weekly and daily surveys on their mental well-being. From this data, my dissertation aimed to broadly: (1) evaluate the validity of past self-report based findings in this field using objective behavioral data, (2) further investigate the strength of associations and discrepancies between self-reports and objective measures of technology use, (3) discover novel objective behaviors related to each of these psychological constructs, and evaluate how well these behavioral features can predict individual differences in these constructs in a machine learning context.

Across 30 hypotheses tested, I show that just 27.2% of hypotheses derived from studies employing self-report assessments, and 75% of hypotheses derived from studies employing behavioral sensing methods, were supported to some degree. Furthermore, I show that commonly problematic technology use scales had relatively weak associations with all objective device usage measures I generated. I also provide further evidence that self-report measuresof specific technology use (e.g. time on smartphones) appear to have at most a moderate association with logged measurements, and suffer from other patterns of systematic biases. From my exploratory analysis, I discovered novel behaviors associated with individual differences in SWL and loneliness that were significant after correcting for false discoveries, and present evidence that SWL, loneliness, and DAS, appear to have a stronger relationship with observable differences in technology use compared to locus of control, impulsivity or narcissism.

Throughout my dissertation, I discuss how integration of behavioral sensing methods into this field provides the opportunity to improve the robustness, precision, reproducibility, and novelty of future findings regarding the relationship between technology use and individual differences in psychological phenomena.

In recent years, political polarization has received increased attention in the United States. Reports suggest that partisan differences have greatly deepened in the past several decades, and that this polarization has had a number of deleterious effects. However, some debate remains as to the etiology of political orientation and political choice—some suggest that political choices can be modeled as utility-optimizing economic actions, while others posit the existence of “moral foundations” that serve as the underpinnings of ideology. We take a quantitative approach to investigating these questions, examining over 80,000 posts from 50 liberal and 50 conservative blogs from the latter half of 2012, around the time of the United States presidential election. In our analysis, we employ Linguistic Inquiry and Word Count, a tool that measures the use of salient linguistic markers in digital texts. The markers examined include function words such as pronouns, articles, and prepositions; researchers have found that the frequency of usage of these words has correlations with a number of phenomena, including age, gender, wealth, and the success or failure of social interactions. Our investigation reveals that some divergence exists between the frequencies of use of these words by liberals and conservatives. By examining these divergences, we find support for a moral foundations approach to understand ideological differences. Our contribution adds to the discourse on the etiology of political choice and probes the ways in which ideological orientation affects written language.

Background

Methods

Findings

Conclusions