Banner Image: We carried out the study in three office rooms, each located in one of the participating campuses. The figure shows the office and setup used at the University of Houston. The experiments in these offices were conducted by personnel trained the same, using identical systems and layouts. During the experimental sessions, the systems continuously imaged the participants’ faces with a thermal and visual camera. An additional visual camera angled down from the ceiling was imaging the participants’ desktop area. The systems were also capturing the screen and keystrokes of the participants’ computer, while two wearable devices were relaying the participants’ physiological signals.
HOUSTON, Nov. 2019 – University of Houston.
Knowledge workers (KW) are `white-collar’ workers whose line of work requires to `think for a living’; examples include scientists, engineers, design thinkers, and academics. KW are widely considered to be the quintessential workforce of the 21st Century. Peter Drucker famously said that `the knowledge worker cannot be supervised closely or in detail; he can only be helped’. Perhaps the best way to help KW is by finding what makes them tick. In this direction, three university laboratories collaborated under funding from the National Science Foundation (NSF) to construct a dataset that would further research in the pains and gains of knowledge workers.
The study design called for 63 volunteers with higher education to carry out a series of typical knowledge tasks and office activities, including writing critical reports, taking breaks, and presenting their findings to management. Special emphasis was paid on exposing the effects of distractions – a plague to modern office work. For this, half the volunteers wrote reports while being regularly distracted by emails (continual mode), whereas the other half dealt with emails only in the context of a dedicated session (batch mode). During the experiment, several instruments recorded unobtrusively data with respect to participants’ physiological stress, displayed emotions, typing habits, and psychometric state.
Stress is believed to be ubiquitous in office environments although identifying its sources and pinpointing its effects have proved elusive. To address this problem, the researchers tracked participants’ stress via both imaging and wearable sensors, they assessed participants’ emotional toll via facial expressions, and quantified participants’ intellectual output via automated tools. For stress, a miniature thermal camera under the participants’ computer monitor recorded facial perspiration; a wireless bioharness worn on the participants’ chest recorded breathing signals; heart rate signals were recorded both on the chest and non-dominant wrist of participants via bioharness and smartwatch, respectively. For displayed emotions, a web camera recorded participants’ facial expressions. Another web camera tucked into the ceiling and angled downwards recorded participants’ hand activity. All these imaging streams and sensor signals were time-synced to facilitate a 360o view of the participants’ internal and external state at every moment. The intellectual output of participants, that is, their reports, were scored by an AI engine. To account for idiosyncratic factors, at specific points during the experimental procedure participants had to fill out online questionnaires about their psychological profile and state.
The research team developed statistical and machine learning methods to curate and validate the collected data prior to releasing them to the scientific community. Byproducts of this data validation process were reported in the journal Scientific Data, providing valuable insights into KW behaviors. Minute fluctuation of facial sweating emerges as the best way to measure stress during knowledge production and handling. Presenting views and findings to management is far more stressful than producing them. Given more relaxed deadlines, many office workers do not write longer reports but spend all the extra time to style them better. Spell checkers – despite their bad rap in the context of texting - save the day in long writings, which would be riddled with mechanics errors in their absence.
`The research that we initiated’, Ioannis Pavlidis leading the University of Houston group said, `exemplifies the evolving nature of human studies’. Gloria Mark, the investigator from the University of California, Irvine complemented that `in this naturalistic experiment, gaining an understanding of how stress unfolds during an office task became possible through unobtrusive sensing’. Ricardo Gutierrez-Osuna, the Texas A&M investigator, underlined that `this was a grand team effort to carry out experimental data collection and validation for the benefit of the scientific community at large’.
Indeed, researchers from around the world are expected to analyze these open data for years to come– a sort of analytic crowdsourcing that maximizes the potential for breakthrough discoveries while ensuring the reproducibility of results. A point in case is that nearly simultaneously with the release of the dataset in the Open Science Framework (OSF), the first analytic study that made use of it was published in the ACM CHI’19 Proceedings – the prestigious forum of human-computer interaction research. The CHI’19 study documents that people with neurotic tendencies are better off when are regularly distracted during a knowledge task, rather than when dealing with distractions in batch mode. This is antithetical to the prevailing one-size-fits-all ergonomic recommendations on the matter, which apparently need to become more nuanced. Continuing analysis of the said dataset is expected to upend more long-held views about knowledge work, thus helping knowledge workers to know themselves better. To extend Peter Drucker’s line of thought, this is the only help KW need in order to unlock their full potential.
FUNDING: This work is supported by the National Science Foundation (grants #1704889, #1704682, and #1704636)
PAPER: To read the paper, go to: Zaman S. et al., Sci. Data. 2019
DATA: To access the dataset associated with the paper go to: https://osf.io/zd2tn//
GITHUB: To get the R code, go to: https://github.com/UH-CPL/Office-Tasks-2019-Methods
BLOG: To read the story behind this research, access the blog of Dr. Pavlidis at: researchdata.springernature.com