*2.4. Procedure*

Figure 2 presents an overview of the methodological procedure employed during the current investigation to collect, process and code the mobile eye-tracking data. After fitting the participant with the mobile eye-tracker, a three-point calibration was undertaken. Subsequent checks for tracking accuracy were undertaken and, if required, calibration was repeated. Data collection took place during the day and not at night, as this could have influenced the data captured [45]. It also took place during fair weather conditions in order to reduce the effect of poor weather, such as rainfall, on participant behaviour within the study streets. A cap was worn by the study participant in order to allow data collection within sunny conditions [44].

Two walked routes were used during the current study, with each route comprising six streets within Sheffield's city centre, UK. Following the initial study setup (see Figure 2, points 1–3), participants were requested to walk one of these two routes. The battery-life of the eye-tracker restricted the duration of data collection to six streets per participant, so two routes were used in order to provide eye-tracking data of street edge visual engagemen<sup>t</sup> from a range of different streets. Both routes comprised non-pedestrianised and pedestrianised urban streets, were discrete, and were devised so that each street had a well-defined start and end point. This was in order to reduce the need for wayfinding, which may have influenced how participants visually engaged with their surroundings.

Before walking along a street of a specified route, study participants were required to read a task card. This introduced an everyday activity for the participant to undertake when walking that street (see Figure 2, point 4). This process took place for each of the six streets of the route walked and provided the opportunity to assess the extent to which everyday activities impacted the distribution of visual engagemen<sup>t</sup> upon the different street edge AOIs. The activities were derived from the on-site observation of peoples' routine behaviours, establishing a degree of real-world and context-specific

validity. In total, six activities were selected across two categories: optional activities (break-time stroll, going for coffee with a friend, window-shopping) and necessary activities (rushing to work, dropping an item off with a friend, walking to the bus). The activities were distributed evenly amongs<sup>t</sup> the streets, with each study participant carrying out an activity once (three optional and three necessary activities across six streets along a single route, see Figure 2, points 4–6).

Data collection across the two routes was undertaken twice; however, eye-tracking data for one street was omitted from the subsequent analyses due to it only having a street edge on one side. This resulted in the data from six non-pedestrianised and five pedestrianised streets being assessed (see Figure 2, points 7–9 and Figure 3 for eye-level images of the study streets). Such a process provided a total dataset of pedestrian visual engagemen<sup>t</sup> with the street edges of 132 walked streets, i.e., 24 study participants walking eleven different streets undertaking different activities that were either optional or necessary.

**Figure 2.** Methods Workflow Diagram.

### *2.5. Data Processing and Coding*

Following the completion of data collection (see Figure 2, points 4–9), each participant's eye-tracking data for each separate street was exported as a single video. VideoCoder [46] was then used to code the gaze dwell duration upon street edge AOIs based upon the knowledge that each video frame indicated the gaze location for a tenth of a second (see Section 2.2 Design for the AOIs used when coding). Processing the data in this way overcame issues with eye-movemen<sup>t</sup> definition, with the raw eye-tracking video output being used prior to the automated classification of eye-movements as either fixations (when the eye is stationary and focused upon a stimulus), or saccades (when the eye is moving and re-adjusting itself) [47,48]. After coding, a log of sequential gaze dwell durations on the AOIs was exported, which could then be used within the subsequent analyses. Depending on the question being posed, these analyses took into account 1) only street edge visual engagemen<sup>t</sup> data in order to compare the percentage of visual engagemen<sup>t</sup> upon a street edge AOI vs. another street edge

AOI (e.g., the amount of visual engagemen<sup>t</sup> with ground floors vs. upper floors), or 2) participant visual engagemen<sup>t</sup> with the entire street in order to assess the percentage of visual engagemen<sup>t</sup> upon a street edge AOI vs. visual engagemen<sup>t</sup> with the rest of the street (e.g., the amount of visual engagemen<sup>t</sup> with ground floors vs. the entire street).

With tracking accuracy in outdoor investigations being typically lower, when compared with laboratory-based eye-tracking, it was anticipated that there would be fluctuations in the data quality. Data loss was generally low, but did vary slightly, resulting in a mean tracking ratio of 93% (range = 68%–99%, standard deviation = 6%). All data captured and coded was used in the following analyses.

**Figure 3.** Eye-Level Images of Study Streets.
