User Study

In the user study different psychological measures sensed by three distinct sensing technologies have been recorded. As a ground truth the subjectively perceived stress level was assessed. In the following, there will be described the study design, the independent and dependent variables, the apparatus used, as well as the sample and procedure.

Study Design Each participant was asked to take part in each of our four conditions lasting 20 minutes in total (5 minutes per condition). In this within-subject design the sequence of conditions was partly randomized following the Latin square distribution but started with the non-stressful- neither mentally, not physically- baseline task. Another constrain was the alternation of mentally and physically stressful tasks to prevent users from exceeding their physical limits.

Three independent variables were used, namely physical activity and mental stress which consisted of two levels each; the triad was supplemented by four different sensing devices recording physiological data throughout the distinct mental and physical strain. The factorial design resulted in four overall conditions which have been performed by all participants. The setup of conditions was inspired by Sun et al. [257] and is illustrated in Figure 3.1.

Variables As independent variables the sensing devices were altered and the amount of physical activity, as well as the level of stress was varied as described subsequently. The physiology-recording hardware was chosen according to their representational function. Hence, the Apple Watch, as a popular smartwatch with fitness capabilities in form of physical activity and heart rate tracking and validated on its performance regarding heart rate errors and correlation with the gold standard device [60, 251, 268] was taken. The second consumer device, the Microsoft Band 2 fitness tracker also relies on optical heart rate recording and is one of the few consumer wearables incorporating skin conductance and skin temperature sensors. To take up another sensing technology, the Polar H7 chest strap as an exemplary device has been used in the study. Being worn by many atheletes and in sports settings due to its convincing accuracy [82], it further bears

the advantage to share sensing data via Bluetooth and send it a mobile phone.

Lastly, the Nexus-10 MK2 by Mind Media, a laboratory measurement instrument has been used. Mostly targeted for biofeedback applications and psychological research it provides the medical standard and was taken as a reference measure.

Physical Activity To observe the devices’ performance in a realistic scenario, the amount of physical activity was designed to be another independent variable. The participants were asked to either walk on a treadmill in their own, physiologically demanding pace resulting in thewalkingcondition or to remain stationary on a comfortable chair accordingly being in the stationarycondition.

Stress Amount The amount of stress perceived by the participants was varied too. For this the participants were asked to perform mental arithmetic tasks (MAT) resulting in thestressfulcondition and, to relax while listening to meditation music being in thenon-stressfulcondition.

Sensing Devices Two wrist-based consumer wearables with optical heart rate technologies (the Apple Watch Series 2 and the Microsoft Band 2³), one chest strap heart rate monitor (Polar H7 chest strap⁴), and a laboratory measurement tool with ECG adhesive electrodes, namely the Nexus 10 kit⁵ were used simultaneously to record elctrodermal activity, skin temperature, and heart rate activity.

Physiological Data From the aforementioned measurement devices heart rate, electrodermal activity, and skin temperature was recorded. There is a lot of previous work providing evidence that these physiological parameters have been shown to be reliable stress indicators [3, 129, 246].

Choice of Physiological Stress-indicating Measures Several studies used physiological measures, i.e. heart rate, electrodermal activity, and skin temperature to detect stress and showed correlation with subjective stress responses [231]. Moreover, the combination of these measures has proved to be a reliable indicator in, e.g. psychology [3, 7], for the development of a non-invasive real-time stress tracking system [159], in a real-world driving tasks to determine the driver’s stress level [107], or for non-invasive stress detection in HCI [12].

3 https://www.microsoft.com/en-us/band

4 https://support.polar.com/e_manuals/H7_Heart_Rate_Sensor/Polar_H7_Heart_Rate_

Sensor_accessory_manual_English__.pdf

5 www.mindmedia.info/CMS2014/en/products/systems/nexus-10-mkii

Baseline Task:

Non-stressful

Stationary Walking Stationary

Fixed Task:

Non-stressful

Walking

Stressful (MAT) Stressful (MAT)

Walking /

/ Stationary

Figure 3.1: Study design depicting the different tasks, as well as the sequence of trials. Participants were asked to perform four trials each, comprising a baseline task in the beginning and a given task between the two alternating trials. For the second and the last task, they were asked to perform mathematical operations while either walking or sitting calmly, alternated according to an counterbalanced order.

Self-reported Arousal, Valence, and Dominance To collect data on the participants’ stress level, they were asked to fill in the widely-used SAM [26].

This self-report measurement tool allows the non-verbal assessment of current affective state, respectively valence (pleasure), arousal and dominance, through pictures. A 9-point rating scale for each dimension according to Bradley’s and Lang’s original work [26] was utilized. Hereby the participants were instructed to place an ’x’ on any of the five figures or between two figures.

This first classical arousal model, similar to Russell’s Circumplex Model of Affect [227], captures tension as one dimension and hence, does not allow differentiations. Thayer [260] suggested that arousal can be further splitted into energetic arousal (ranging from wide-awake to tired) and tense arousal (nervous to calm). This fine graded characterization was taken up and Schimmack’s and Grob’s [234] recommendation to add two additional questions was followed for the study: a 5-point self-rating Likert-item for each dimension assessing tension and wakefulness [123, 222] was added.

Apparatus The stimulus material consisted of Mental Arithmetic Tasks (MAT) which had been proven to elicit stress in participants [18] and also influence physiological measures [104, 160, 245, 262]. To ensure that the physiologically observable stress was also subjectively perceived by the participants, a visual countdown and feedback (both visual and auditory) was added. Setz et al. [246]

showed that this is a valid method to further increase stress. Feedback was always given after the participant’s response. While for correct answers the screen turned green and the word "Correct" was displayed, false answers or time outs were signified with either "False" or "Time out" being shown on a red screen and a buzz sound. For each of the calculations, addition and subtraction of two-digit numbers ranging from 0-100 and including negative solutions, the participants were given six seconds each. A timer representing the time left for each task at the bottom of the display was provided. According to the study design, the relaxing task consisted of listening to meditation music⁶. These tasks were presented on a 60-inch display being positioned on a desk in front of the participants. The study setup was inspired by Vlemincx et al. [267]. Regarding the performance of physical activity, participants were asked to walk for five minutes on a treadmill (model: ProFitness Sierra motorized) for the walking task.

Participants and Procedure In total, 24 participants for the study were acquired via university mailing lists, leaflets and personal recruitment campaigns.

For the final data analysis, one pilot participant who had volunteered for a test trial of our laboratory setup and two other participants due to technical problems during the data collection were excluded . Hence, 21 participants with a mean age of 28.9 years (SD=4.5) remained; among them were eight females and 13 males. Before the recruitment, strict exclusion criteria were set up, namely no participants being taken who had been diagnosed with any heart conditions, mental illnesses or learning disabilities. Participants were also asked to guarantee that they did not suffer from alcohol and/or drug addiction. Before the experiment started, they were asked to refrain three hours from caffeine. For compensation each participant received 15 British pounds. Prior to the study, the participants were introduced to the experiment environment and briefly informed about the study background as well as the sensor placement on the body. They were asked to sign the consent form and to fill in an initial assessment consisting of demographic questions, self-reported fitness level, and smoking behavior as inquired in Weitkunat et al. [271].

After they were given a short introduction to the treadmill and the mental arithmetic task was explained, the participants were asked to put on the chest-worn

6 As meditation music the song number 14 from the album ’72 Ambient Meditations’ was used

ECG sensors (Nexus 10 ECG with pre-gelled, disposable electrodes) and the Polar H7 chest belt. Hereby a sheet with visual material from the manufacturers and instructions on the correct sensor placement was provided to ensure proper sensor fit was provided. Those illustrations where discussed with the participants, before they proceeded. The sensor was attached with medical tape on the forearm because it should be comparable to the values provided by the Microsoft Band and therefore was attached close to it’s location. Instructions were provided how to place the Polar H7 chest belt around the chest and close the clasp after moistening the plastic parts at the inside of the strap. Participants were also instructed to already put the cables for the skin temperature and skin conductance through the sleeve of their right arm, to prepare for the following steps. The participants were given privacy behind a screen to perform these actions. The correct placement of the sensors and the correctness of the ECG signal was later checked on the experimenter’s computer. After the chest sensors were placed, the experimenter placed the skin conductance sensors on the fingers and attached the skin temperature sensor on the participant’s forearm using medical tape. Next, the participants were helped to place the wrist-worn devices (Apple Watch and Microsoft Band 2). Before the study started, the correct data transmission and validity of the signals for all sensors was initially checked. Each procedure starting with the recording of the baseline. In this condition all participants were asked to remain seated for five minutes listening to meditation music via wireless headphones. Then, each participant was assigned the following conditions in counterbalanced order, alternating between walking and stationarywhile mental arithmetic tasks should be performed. Between the two stress-inducing trials, there was a fixed task requiring to walk while listening to meditation music via wireless headphones. This sequence of conditions is illustrated in Figure 3.1. To assess our dependent variables, the SAM questionnaire including single-items on wake/tense arousal and perceived stressfulness of the task after each trial (including the baseline) was administered. The entire procedure took approximately 1.5 hours in total.