While all these solutions attempt to deal with the effect of walking, our goal in this chapter is to understand users’ performance in order to build new solutions that can bridge the gap between SIID and HIID, namely in tremor-induced impairments.
Figure 4.3: Participant following the pacesetter while entering text.
Next, participants filled in a pre-questionnaire about their demographic data and mobile phone usage (see Appendix A4 for details). Participants were then informed about the experiment and how to use our evaluation application.
We evaluated the participants’ performance in three conditions: sitting, walking at average human pace (2 paces per second), and walking at 65% of average human pace (1.3 paces per second) (Barnard et al., 2005). The experiment was conducted in an indoor test track built-up at IST-Tagus Park campus (Figure 4.3). In the sitting condition, participants sat at a desk in a controlled and quiet environment. We instructed them to remain seated until they completed all text-entry tasks. In both walking conditions, we asked participants to follow a pacesetter while entering text. Although other designs could be chosen (Kane et al., 2008b), we opted to keep a fixed pace rather than measure it as a dependent variable in order to ensure a comparable level of walking demand across trials. The experimenter instructed participants to stay within 2 meters of the pacesetter as he walked. If the participant fell behind the pacesetter by more than 4 meters, the experimenter logged a walking deviation for that trial. The pacesetter carried a mobile phone, which gave him feedback through vibration, about the intended pace. Also, before each mobility condition participants had a 5 minute practice trial to get used to the pace and text-entry task.
For each mobility setting, they were asked to enter text with 3 hand postures (chosen randomly), always using their thumbs: portrait/one-handed, portrait/two-handed, and landscape/two-handed. For each condition participants copied seven different sentences (the first two sentences were practice trials), displayed one at a time, at the top of the screen (Figure 4.4 - right). Copy typing was used to reduce the opportunity for spelling and language errors, and to make error identification easier. Participants were instructed to type phrases as quickly and accurately as possible.
We wanted to elicit natural typing behaviors and did not want participants to be concerned
Figure 4.4: Text-entry application and virtual keyboard on portrait model (left), and HTC Desire (right).
with the accuracy of their input. Thus, we followed a similar approach to Gunawardana et al. (Gunawardana et al., 2010) and Goel et al. (Goel et al., 2012), and created a keyboard in such a way that error correction was not available. On the other hand, if the prototype had a delete key it might introduce correcting strategies, which might vary across participants and upset the naturalness of the data. Participants were told that they could not correct errors and were instructed to continue typing if an error occurred. Once participants finished entering each sentence, they pressed the ’next’ button to receive a new sentence. When the seven sentences were entered, we asked participants to perform the same tasks on a new mobility condition, until they performed all mobility settings.
Each participant entered a total of 63 different sentences. These sentences were extracted from a written language corpus, each with 5 words, an average size of 4.48 characters per word, and a minimum correlation with the language of 0.97 (see Apendix A1). Both sentences and mobility conditions were chosen randomly to avoid bias associated with experience.
4.2.4. Apparatus
A HTC Desire with a capacitive touchscreen (see Figure 4.4 - left) was used in the user study. A QWERTY virtual keyboard was used to simulate a traditional touch keyboard, where each key was 10x10mm on landscape mode, and 7x10mm on portrait mode. Neither word prediction nor correction was used. Acceleration data was captured through the device’s accelerometer for later analysis.
Regarding the pacesetter, he also had a mobile device, which gave him feedback through vibration, so he could maintain a steady pace.
4.2.5. Dependent Measures
The performance during text-entry tasks was measured by several quantitative variables:
Words Per Minute (WPM),Minimum String Distance (MSD) error rate, and character- level errors (substitutions, insertions, and omissions). Qualitative measures were also gathered in the end of the experiment by debriefing each participant. Walking errors, such as slowing down and stopping were recorded by the experiment supervisor.
We also gathered the motor demand (acceleration) of mobility conditions. This was achieved through the device’s accelerometer sensor. Thus, with this measure we were able to objectively characterize the motor demand of all mobility conditions for further analysis.
4.2.6. Design and Analysis
We used a within-subjects design where each participant tested all mobile conditions. For each condition each participant entered 5 test sentences, resulting in a total of 45 sentences per participant. In summary the study design was: 22 participants x 5 sentences x 3 hand postures x 3 mobility settings (1 seated + 2 walking conditions) = 990 sentences overall.
Shapiro-Wilkinson tests of the observed values for WPM, MSD error rate, and types of errors showed to fit a normal distribution for all conditions. Therefore, a two-way repeated- measures ANOVA (mobility x hand posture) was used in further analysis. Greenhouse- Geisser’s sphericity corrections were applied whenever Mauchly’s test of sphericity showed a significant effect. Pairwise Bonferroni corrected t-tests were used for post-hoc tests.