Using Bluetooth Beacon Technology to Reduce Distracted Pedestrian Behavior: A Crossover Trial Study Protocol

Abstract

Objective.

Over 6,400 American pedestrians die annually, a figure that is currently increasing. One hypothesized reason for the increasing trend is the role of mobile technology in distracting both pedestrians and drivers. Scientists and policymakers have attended somewhat to distracted driving, but attention to distracted pedestrian behavior has lagged. We will evaluate Bluetooth beacon technology as a means to alert and warn pedestrians when they approach intersections, reminding them to attend to the traffic environment and cross streets safely.

Methods.

Bluetooth beacons are small devices that broadcast information unidirectionally within a closed proximal network. We will place beacons at an intersection frequently trafficked by urban college students. From there, the beacons will transmit to an app installed on users’ smartphones, signaling users to attend to their environment and cross the street safely. A crossover trial will evaluate the app with 411 adults who frequently cross the target intersection on an urban university campus. We will monitor those participants’ behavior over three distinct time periods: (a) 3 weeks without the app being activated, (b) 3 weeks with the app activated, and (c) 4 weeks without the app activated to assess retention of behavior. Throughout the 10-week period, we will gather information to evaluate whether the intervention changes distracted pedestrian behavior using a logistic regression to estimate the likelihood of key behavioral outcome measures and adjusting for any residual confounding. We also will test for changes in perceived risk. The trial will follow CONSORT statement guidelines, as modified for crossover design studies.

Over 6,400 American pedestrians die annually,¹ a figure that is currently increasing. One hypothesized reason for the increasing trend in pedestrian injuries and deaths is the role of mobile technology in distracting both pedestrians and drivers. Existing behavioral interventions to reduce distracted pedestrian behavior are few. This research will evaluate the novel strategy of using Bluetooth beacon technology to alert and warn pedestrians when they are approaching dangerous intersections, reminding them to attend to the traffic environment and cross the street safely rather than engaging with mobile technology.

Significance of Problem

The rate of pedestrian deaths is increasing in the United States, approaching 6,500 deaths in 2017, the latest data available.¹ About 11% of those killed were young adults ages 17-25. Non-fatal, medically-attended pedestrian injury incidents number almost 200,000 annually, and also show increasing recent trends.¹

Making causal assumptions from injury trends is risky, but the growth of pedestrians distracted by mobile devices is hypothesized to play a significant role in recently-increasing US pedestrian injury rates.^2–4 Pew Research data suggest US mobile phone use increased quickly and steadily over the past 15 years, and there are now more smartphones in the US than people.⁵ Smartphone penetration to young adults in the United States is nearly 100%.⁵

Supporting the possibility that distraction interferes with safe pedestrian behavior, cognitive science research suggests multitasking – attempting to complete two cognitively complex tasks simultaneously – causes attention to and performance on one or both tasks to decrease.⁶ Use of mobile devices requires cognitive load, reducing the individual’s ability to focus cognitive effort on the street-crossing task. Beyond cognitive distraction, some types of distraction (e.g., texting, internet browsing) diminish visual attention on the street environment and others (e.g., listening to music) diminish aural attention. In all cases, reduced cognitive, visual, and/or aural attention reduces pedestrian safety.^7–12

Bluetooth Beacon Technology

Bluetooth beacons are very small (about the size of a dime) and inexpensive (~$20 range) devices that broadcast information unidirectionally (beacon to smartphone) within a closed proximal network. We will place multiple beacons in discrete locations at traffic intersection frequently trafficked by urban university students. From that location, the beacons will transmit to an app installed on users’ smartphones, signaling them to attend to their environment as they approach the intersection and cross the street safely.

Previous Research

Distracted pedestrians take greater risks than undistracted pedestrians.⁴ Background research may be divided into observational vs simulated research.

Initial observational studies found pedestrians talking on phones^13–14 or distracted by activities including eating, drinking, smoking, or talking¹⁵ took greater risks crossing streets than those who were undistracted. More recently, Thompson and colleagues observed urban Seattle pedestrians and reported about 30% were distracted and distracted pedestrians – especially those using their phones for text messaging – took longer to cross and crossed in riskier ways.¹² Similarly, Basch and colleagues found 42% of New York City pedestrians who crossed illegally (i.e., ‘don’t walk’ sign was lit) were distracted.¹⁶ Others report similar findings in urban areas^17–18 and college campuses¹⁹.

Research in simulated environments supports observational data. Stavrinos and colleagues reported diminished pedestrian safety among children¹¹ and adults on phones, listening to music, texting, and internet-browsing.^7–10 Results vary somewhat across studies, but effect sizes range from d = 0.35 to 0.72 for experimental effects of adult distraction on hits/close calls in simulation, and larger effects are reported for secondary measures such as attention to traffic and time before vehicle contact while crossing.^7,9–10 In secondary analysis of one study’s data, 60% of young adults experienced at least one virtual hit or close call (<1 sec from impact) while crossing the street and texting in a 10-crossing trial with light traffic; exposure to heavier traffic and more trials would likely yield higher rates.⁹

Published reports of behavioral interventions to reduce distracted pedestrian engagement in traffic are few. Attempts to warn pedestrians about distracted pedestrian behavior with stenciled warnings at the curb-cut show mixed results, especially over time.^19–20 Our own program to offer experiential learning about the risk of texting while immersed in a virtual pedestrian environment, plus widespread social and traditional media on an urban college campus, yielded some change in self-reported behavior among a subsample exposed to the experiential simulation, but no changes in community-wide observed behaviors.²¹ Policymaking has been attempted in a few jurisdictions—including most prominently in Honolulu, Hawaii—but not subject to rigorous empirical investigation. Thus, there is limited empirical evidence of successful distracted pedestrian prevention programs.

The Present Study

The present study will evaluate StreetBit, an app we created to alert distracted pedestrians that they are approaching an intersection and should attend to oncoming traffic. StreetBit is downloaded to smartphones and functions through alerts from Bluetooth beacons, offering a mostly-passive intervention that intrusively informs pedestrians when they are approaching a street corner while distracted at the moment of risk. The message is delivered in visual (warning on phone screen), tactile (vibration) and aural (verbal message) formats, flashing over existing apps both visually and aurally as the user approaches the intersection.

StreetBit differentiates smartphone users who are driving through the intersection in a vehicle from those walking by capturing speed and trajectory of travel. It successfully captures pedestrians approaching the intersection from any angle using triangulated assessment of movement from multiple beacons and successfully determines context of user activity through assessment of active apps on the phone. User privacy is protected through unidirectional communication (from beacon to phone only) and user acceptance of data storage regulations that replicate pervasively-used mobile phone apps such as Google Maps and web browsers. The app functions on both Android and iOS platforms, running in the background and functioning only when users are near the intersection beacons.

We will conduct a crossover trial to evaluate the efficacy of StreetBit in changing distracted pedestrian crossing behavior. Crossover trials offer an advantage over other experimental research designs in that each study participant serves as his or her own control. The primary limitation of crossover trials is the influence of the first exposure on the second. Replicating features of a randomized trial, however, we will mitigate this limitation by collecting baseline data first, prior to the app initiation, and then initiating the app during an intervention phase. The study is registered at clinicaltrials.gov, NCT03604497.

Research Site.

The intervention will take place at the corner of 14^th Street South and University Boulevard, a busy intersection in the heart of the University of Alabama at Birmingham (UAB) campus in Birmingham, Alabama. UAB is an urban university where students, staff, and faculty cross streets frequently. In recent research with 138 UAB undergraduates, students self-reported walking an average of 10.63 blocks per day (Schwebel, unpublished data).

To offer a sense of pedestrian traffic volume, the intersection of 14^th Street South and University Boulevard had 916 pedestrian street-crossings/daytime hour in recent sampling on 10 occasions over three weekdays (Schwebel, unpublished data). Vehicle traffic through that intersection is also heavy, with sampling on 10 occasions over three days yielding traffic volume on University Boulevard at 1480 vehicles/hour and on 14^th Street at 548 vehicles/hour.

Participants.

We will recruit 411 adults ages 17 and over who cross streets on the UAB campus regularly (at least four times each week, selected to capture students who have class two times/week on the opposite side of the intersection from their dormitory; we expect most of the sample will cross the intersection considerably more often), are willing to install the app on their smartphone, and can communicate fluently in English. We expect most participants will be UAB students but will not exclude others who meet study criteria, such as university staff members. Participants will be reimbursed for their time.

General Protocol.

Participants will be recruited through social and traditional media on campus, word-of-mouth, flyers, and targeted recruitment in classrooms and orientation sessions. Following consenting, participants will complete brief questionnaires (detailed below) and then work with a researcher to download the app onto their smartphone. The app will remain on the participant’s phone for the subsequent 10 weeks, following a standardized schedule of 3 weeks of no alerts (baseline of typical behavior at street corners), 3 weeks of app functioning to provide alerts at corners, and then 4 weeks of no alerts (assess retention of learned safe behavior). Following the 10-week period, participants will return to the lab to complete follow-up questionnaires and receive compensation.

Questionnaires.

We will assess basic demographics, pedestrian behavior, and perceptions about risk from distracted pedestrian behavior using a 25-item survey (30 items on follow-up) addressing the following four topics:

• Demographic Information. Administered only pre-intervention, to assess age, gender, race/ethnicity, position at the university (e.g., student, faculty, staff, not affiliated), and, for students, year/status in school

• Pedestrian Behavior Questionnaire. Based on questionnaires used previously,^7,9–10,21 the instrument has two parts. First, a walking diary will report typical street-crossing behavior on various days of the week. Second, a set of items will assess self-reported walking habits while distracted, both on campus and in other high-risk areas such as downtown and nightclub districts.⁷

• Perception of Risk. Also adapted from previous research,^7,21 this measure will assess perceived vulnerability to pedestrian injury while crossing streets distracted in a range of environments (e.g., campus, downtown, nightclub, neighborhood). The instrument addresses perceived risk walking while distracted in four ways (talking on the phone, texting, using apps, listening to music).

• Post-Intervention Survey. Administered only post-intervention, this survey will assess perceptions about the app’s functionality, usability, and perceived value.

Behavioral Outcome Measures.

Behavioral outcomes will be collected electronically via the app, which will record participant behavior during and following contact with beacons placed across campus. Feasibility and validity of electronic data collection will be iteratively and rigorously pilot-tested prior to implementation. Assessment will record a wide range of behaviors, ranging from the user’s interactions with the phone (e.g., modifying phone’s position to look at it or taking it out of pocket/purse, change in the user’s movement such as stopping, slowing down, stopping or pausing music or other apps while crossing the road). Standard features on smartphones such as front and rear cameras, gyroscopes, accelerometers, light sensors, GPS locators, and positioning with respect to the beacons will be used to gather these electronic data. Once collected, the data will be uploaded from smartphones to a secure server on a daily basis. Data will only be collected when users are in close proximity to the beacons, but large amounts of data will be collected nonetheless. Data will be codified on the backend server following daily data submission from the smartphones. The following comprise the primary outcomes:

1. Activity immediately following contact with a beacon, categorized as: no phone use phone call, texting, browser actively engaged within, app actively engaged within, and/or music playing

2. Phone activity 15 seconds following contact with beacon: no phone use, phone call texting, browser actively engaged within, app actively engaged within, and/or music playing

3. User activity during 15 seconds following contact with beacon: walk across street toward other beacons, walk but not across street (e.g., down sidewalk other direction), stay still and look at phone (phone moves upward), stay still and not look at phone (for example, engaged in a conversation)

4. Activity in response to visual warnings (intervention phase of study only, when applicable): ignored, swiped to remove, looked at but not immediately swiped to remove

5. Taking the above four criteria together, categorization into: pedestrian appears to stop distracted pedestrian behavior after contact with beacon, pedestrian does not appear to stop distracted behavior after contact with beacon, or no distraction present upon approaching intersection

The above five criteria will be assessed every time a participant crosses the intersection marked with beacons. Much as these electronic data collection strategies are somewhat unusual in public health, traffic engineering, and social science research, such assessment of human behavior is comparatively routine in computer science research.²²

Statistical Analysis Plan.

Our primary hypotheses consider whether the intervention changes distracted pedestrian behavior. We also will consider whether risk perceptions change. We plan to first conduct analyses among all three time periods at once, and then in post hoc analyses perform pairwise comparisons between each period (i.e., pre-intervention to intervention, intervention to post-intervention, and pre- to post-intervention) to determine between which study periods differences exist. We will conduct bivariate data analysis using paired t-test (Wilcoxon signed ranks for non-normally distributed data) or repeated measures ANOVA and either McNemar’s Q (for 2x2 contingency analyses) or Bowker’s test of symmetry (for 2xn contingency analyses) for continuous and categorical variables, respectively, due to the crossover trial design. Conditional or General Estimation Equation logistic regressions will estimate the likelihood of behavioral outcome measures, adjusting for any residual confounding (though we anticipate very little as individuals serve as their own controls). In secondary analyses, we will examine whether the effects of the app on distraction prevalence is differential by demographic characteristics such as age and gender. We also will consider descriptive reports from participants concerning the app’s usability, functionality, and perceived value.

Sample Size.

We based sample size estimates on our prior work, which reported an overall self-reported distraction rate of 33% at baseline and a 40% decreased odds of distraction due to texting and talking following the intervention.²¹ Using these estimates, we estimate 370 individuals will need to be recruited to achieve 80% power to detect a difference in the prevalence of distraction. Accounting conservatively for a 10% dropout rate, we will recruit 411 individuals.

Power analyses were conducted conservatively, but if the proportion of data missing is unexpectedly large, we will utilize multiple imputation methods to fill missing values and assess sensitivity of results to the imputation. We anticipate some attrition from the study and intention-to-treat analyses will be conducted.

Study Limitations.

Our research design provides a thorough initial evaluation with a sophisticated design, but is limited to a single location. Generalizability will be tested in subsequent, larger trials. The software may raise privacy concerns but we will reassure users of the unidirectional nature of the beacons (they send information to smartphones but collect none), and the fact that beacon technology is less invasive and tracks less information than many apps currently installed on most smartphones (e.g., Google Maps). Our primary outcome measures will be collected through electronic data collection using participants’ smartphones. Extensive pilot-testing will be conducted to ensure validity of those outcomes. However, such electronic data collection is arguably more accurate than self-report or behavioral observation measures, both of which are subject to various measurement risks and biases.

Future Directions.

If this program proves successful, we anticipate extending it to broader geographic areas and demographic populations. Critics may wonder if the intervention could ever work, whether users would ever accept it, and what the scalability is. Empirical research is needed, but we use as a model the history of seat belt adoption in the US. In the 1960s, seat belt use was rare. A combination of factors moved society to today’s 86% usage.²³ One factor was behavioral: Auto manufacturers installed buzzers and alarms to sound intrusively when seat belts were unfastened. This loosely parallels our proposal to intrusively provide buzzers and warnings in smartphones when distracted pedestrians approach intersections. Another factor in seat belt usage was policy: States mandated seat belt use and penalized motorists who disobeyed. Engineering of placing beacons in locations like signal-posts (e.g., attached to walk/don’t walk signs) is basic, and costs are comparatively small. With rigorous empirical study of efficacy, societal costs to warn distracted pedestrians through a beacon system would far outweigh costs of pedestrian injuries. Installation of countdown warnings for pedestrians in many locations, and increasing use of audible sounds to assist vision-impaired pedestrians offer existing models of altered US pedestrian crossings. Initiatives to pre-install “Do Not Disturb While Driving” features on Apple iPhones indicate industry motivation to reduce injury risks to users through pre-installed phone features.²⁴ We view this research as a first step to a potentially transformative intervention that reduces distracted pedestrian behavior and saves lives.