3.3.2. Study Procedures

During the main part of the study, the subjects were asked to secure the phone on their body in a comfortable area (e.g., on the shoulder or on the hip) during their routine runs with the RunningCoach app. In addition, the subjects were asked to wear the heart rate chest strap that was provided to them, which connects to the app via Bluetooth and sends the data in real time. The subjects were not specifically asked to use the app during each run but rather to use it on their own terms. The reason subjects were not instructed to use the app during each run is that some recruited subjects are competitive runners who preferred to use a smartphone only during a part of their weekly training routine.

During each run, the app collects information about the runners' estimated energy expenditure, cadence, speed, heart rate from the chest strap, and total distance covered. In addition to these estimates, the app collects the following two variables: (i) whether or not the screen light is on; and (ii) the battery level. Before and after each run, the app collects single estimates of the heart rate using two different vision algorithms, one from a video of the subject's face and another from a video of the subject's index finger [11]. These algorithms were previously only validated under controlled conditions. Since an external heart monitor was used in the study, we used this opportunity to get an insight on the usability of the implemented heart rate measurement algorithms in the field. Note that this was not a controlled validation and should therefore be treated as exploratory only. Table 2 summarizes the types of data collected before, during and after the run.

**Table 2.** The data types collected about every run.


After each run, subjects were asked to fill a post-run survey inquiring about the run. The questions asked in the post-run survey were: (i) "How tired were you on a scale of 1-5 where 3 is your typical level of fatigue after long runs prior to using the app, 5 is very tired and 1 is least tired?" (ii) "After viewing your run data, were any of the measurements inaccurate to the best of your assessment? (Choose all that apply from Speed, Cadence, Heart Rate, Energy Expenditure, Distance);" (iii) ''If you selected any of the choices in the previous question, please explain;" and (iv) "Please provide any other comments regarding your experience using the app." Figure 2a depicts an example screen showing some of the run statistics after the run, and Figure 2b depicts an example of a question from the post-run survey as displayed in the app. The post-run survey provides information about (i) the perceived accuracy of the system; (ii) the usability of the system; and (iii) how hard the training regimen is pushing the runners in terms of performance. The post-run questions about the perceived accuracy of the app's collected data are shown to the subject after the run statistics are presented (e.g., Figure 2a).

In the process of designing the post-run survey, the 1 to 5 scale of fatigue was selected for the following reasons. The adopted scale is a reduced version of rating-of-fatigue (ROF), which is a 10-point scale designed to measure level of fatigue [42]. Other seemingly relevant measures have been studied in the sports literature, including Borg's perceived exertion scale [43]. Borg's scale is not a good fit for our purposes because it is designed to capture subjective exertion. Some researchers argue that perceived exertion, the subjective experience of how hard a physical task feels, is different from perceived fatigue and should not be used to measure perceived fatigue levels [42]. Moreover, Borg's scale is designed to follow the heart rate of the subject by multiplying it by 10. Since we are collecting heart rate data, Borg's score would not provide additional information, and therefore, a reduced fatigue scale that is similar to ROF was selected for our purposes.

**Figure 2.** (**a**) A screen showing some of the run-statistics after a run; and (**b**) an example of a question in the post-run survey [11].

Subjects were asked to perform the aforementioned procedure for a period of 3 months. By the end of that period, the subjects answered an exit acceptability and privacy survey about the system and the study.
