Abstract
Objective:
We examined how sending mobile-device warnings to texting pedestrians when they initiate an unsafe road crossing influences their decisions and actions.
Background:
Pedestrian texting has been identified as a key risk factor in pedestrian–vehicle collisions. Advances in sensing and communications technology offer the possibility of providing pedestrians with information about traffic conditions to assist them in safely crossing traffic-filled roadways. However, it is unclear how this information can be most effectively communicated to pedestrians.
Method:
We examined how texting and nontexting pedestrians crossed roads with continuous traffic in a large-screen, immersive pedestrian simulator using a between-subjects design with three conditions: texting, warning, and control. Texting participants in the warning condition received an alarm on their cell phone when they began to cross a dangerously small gap.
Results:
The results demonstrate the detrimental influence of texting on pedestrians’ gap selection, movement timing, and gaze behavior, and show the potential of warnings to improve decision making and safety. However, the results also reveal the limits of warning texting participants once they initiate a crossing and possible overreliance on technology that may lead to reduced situation awareness.
Conclusion:
Mobile devices and short-range communication technologies offer enormous potential to assist pedestrians, but further study is needed to better understand how to provide useful information in a timely manner.
Application:
The technology for communicating traffic information to pedestrians via mobile devices is on the horizon. Research on how such information influences all aspects of pedestrian behavior is critical to developing effective solutions.
Introduction
Pedestrian injuries and fatalities caused by collisions with motor vehicles are a major public safety concern in the United States. In 2015 alone, there were 5,376 pedestrian fatalities and 70,000 pedestrian injuries involving motor vehicle collisions (National Center for Statistics and Analysis, 2017). Distraction has been identified as a contributing factor to pedestrian–vehicle crashes, with studies showing that mobile devices, such as cell phones, impair pedestrian and driver attention (Ferguson, Xu, Green, & Rosenthal, 2013; Strayer, Drews, & Crouch, 2006; Thompson, Rivara, Ayyagari, & Ebel, 2013; Violano, Roney, & Bechtel, 2015). An important issue is devising ways to ameliorate the harmful effects of texting on pedestrian road crossing. This paper presents a study examining how texting pedestrians respond to warnings on a cell phone when they initiate a dangerous crossing in a pedestrian simulator (Figure 1).
A texting pedestrian waiting to cross a virtual street in the pedestrian simulator.
Researchers have used both observational and controlled laboratory studies to examine how distraction from a mobile device affects pedestrian behavior. This work has demonstrated that people exhibit riskier pedestrian behavior when texting or talking on a cell phone (Lin & Huang, 2017; Nasar & Troyer, 2013; Neider, McCarley, Crowell, Kaczmarski, & Kramer, 2010; Schwebel et al., 2012; Stavrinos, Byington, & Schwebel, 2011). For example, distraction from mobile-device use results in more unsafe crossing behaviors, such as walking against traffic lights or failing to look both ways before crossing (Bungum, Day, & Henry, 2005; Nasar, Hecht, & Wener, 2008). Other observational work has shown that pedestrians using a mobile phone cross the roadway more slowly, thereby increasing their exposure to traffic (Hatfield & Murphy, 2007). Schwebel and colleagues found that texting participants were more likely to experience collisions or close calls than nontexting participants in a virtual road-crossing task (Byington & Schwebel, 2013; Stavrinos, Byington, & Schwebel, 2009). Together, these studies clearly show that distraction from mobile-device use increases the likelihood of collisions between pedestrians and vehicles.
An important question is how to mitigate the harmful effects of mobile-device use on pedestrian safety. The best solution is to advise pedestrians to refrain from using their mobile devices while crossing roads. However, as with mobile-device use while driving, many pedestrians will not heed this advice. Another approach to improve safety for texting pedestrians is to integrate them into the roadway communication loop by incorporating connected-vehicles technology into smartphones (Anaya, Merdrignac, Shagdar, Nashashibi, & Naranjo, 2014; Creaser, Rakauskas, Ward, Laberge, & Donath, 2007; Wang, Cardone, Corradi, Torresani, & Campbell, 2012). The goal is for vehicles and phones to exchange information about their positions and movements. This vehicle-to-pedestrian (V2P) communication would make pedestrians “visible” to drivers even when occluded by an object or in the dark. Researchers have developed mobile phone apps that exchange information with nearby vehicles and send collision warnings to both the driver and the pedestrian (Hussein, Garcia, Armingol, & Olaverri-Monreal, 2016; Wu et al., 2014). Despite such progress in developing the technology to support V2P communications, little is known about how to most effectively present information about roadway conditions to pedestrians. A particularly pressing problem is whether pedestrians will trust and attend to the information delivered through a mobile device, especially while they are texting.
Recently, Rahimian et al. (2016) used an immersive pedestrian simulator to examine how texting pedestrians responded to “permissive” alerts sent to their cell phone about when it was safe to cross a roadway. Participants physically crossed a stream of continuous traffic while texting on a cell phone. A countdown clock and an auditory signal communicated when a crossable gap was about to arrive at the crosswalk. Their gap decisions and crossing actions were compared with those of a texting group who crossed the road without alerts and a control group who crossed the road without texting. Participants in the alert and control groups made more conservative and discriminating gap choices than did participants in the texting group. However, using the alert system was not without costs. Notably, texting participants who received alerts spent much less time looking toward the traffic than did texting participants who did not receive alerts or the control group (who spent the most time looking at traffic). This outsourcing of cognitive processing may lead to reduced situation awareness, which could result in a reduced ability to react to unexpected events or technological failures.
A complementary approach to assisting safe road crossing is to warn texting pedestrians when they attempt to cross an unsafe gap. We call these “prohibitive” warnings because they send a signal meant to prohibit a pedestrian from crossing a gap that is likely to result in a collision. Such alarms may prevent or reduce dangerous crossings without diminishing attentiveness to traffic conditions (as found with permissive alerts), because the warnings do not identify crossable gaps but instead warn the pedestrian when he or she is initiating an unsafe crossing. Because warnings do not aid in gap selection, texting pedestrians should allocate attention to traffic in similar ways with and without warnings.
Here, we examined how texting pedestrians respond to a warning sent when they begin to cross a dangerously small gap. Our goals were to assess (a) the effects of texting on pedestrian road-crossing behavior and (b) the effectiveness of warnings in reducing unsafe road crossing in texting pedestrians. We predicted the following:
Texting will reduce attention to traffic and impair road-crossing performance.
Warnings will ameliorate the detrimental effects of texting, leading to a larger average gap size and fewer risky crossings and collisions.
Texting pedestrians who receive warnings will spend a similar amount of time attending to traffic as texting pedestrians who do not receive warnings.
Method
Experiment Design
We used a between-subjects design with three conditions: texting, warning, and control. In the texting condition, participants received and responded to text messages throughout the road-crossing session. The warning condition was identical to the texting condition except that participants also received an auditory alarm from a cell phone when they began to cross a gap that was classified as unsafe. In the control condition, participants held a cell phone throughout the road-crossing session but did not text or receive alerts.
Apparatus
Our virtual environment consisted of three screens placed at right angles relative to one another, forming a three-walled room (4.33 × 3.06 × 2.44 m). Three DPI MVision 400 Cine 3D projectors rear-projected high-resolution, textured graphics in stereo onto the screens. An identical projector front-projected high-resolution stereo images onto the floor (4.33 × 3.06 m). Stereo sound was used to generate spatialized traffic sounds. Participants wore Volfoni ActiveEyes stereo shutter glasses and a helmet with reflective markers. Reflective markers were also mounted on the cell phone. An OptiTrack motion capture system was used to determine the position and orientation of the cell phone and the participant’s head based on the marker locations viewed from 17 Flex 13 cameras. The participant’s eye point was estimated from the head data and used to render the scene for the participant’s viewpoint. The virtual environment software was based on the Unity3D gaming platform. In-house code generated traffic and recorded the positions and orientations of the vehicles, the pedestrian, and the cell phone for later analysis.
Traffic Generation
A stream of traffic traveled from left to right on a one-lane road (Figure 2). Vehicles were generated from behind a house on the left-hand side of the road, passed through the screen volume, and then disappeared behind a house to the right. The road initially curved and then approached the participant along a straight section of roadway that was perpendicular to the left screen. The length of the visible portion of the road was selected so that the tail vehicle in the next gap always appeared before the lead vehicle passed the participant. Thus, participants could always see the entire gap before they began to cross the road.
Perspective view of the roadway environment.
Vehicles were timed so that the temporal gap between vehicles at the point of crossing (i.e., the time between the moment the rear of the lead vehicle crossed the center line and the moment the front of the tail vehicle crossed the center line) was one of five preselected gap sizes (2.5 s, 3.0 s, 3.5 s, 4.0 s, or 4.5 s). Note that the temporal and spatial gap between two vehicles traveling at different speeds changed continuously as the vehicles approached the intersection. To create moderately dense traffic, small gaps occurred more frequently than large gaps according to the distribution shown in Figure 3. On each trial, the simulator generated a sequence of vehicles for which each gap was randomly drawn from the 13 given gaps, and vehicle speed was randomly set to either 40.23 or 56.33 km/h.
Distribution of gap sizes.
Texting
An Android messenger application was developed in-house and installed on a cell phone. Messages sent to the participant’s cell phone were shown as coming from our simulator, “Hank.” Hank asked a sequence of questions, waiting for the response to each question before sending the next message. Participants were asked to respond to each question with a single reply. Participants were notified of the arrival of a new message by a 0.5-s vibration of the cell phone.
Cell Phone Warning
We developed a warning system to (1) detect when a participant began to initiate a crossing motion, (2) determine whether there was sufficient time to safely cross through the chosen gap, and (3) send an alarm to the participant’s cell phone if the chosen gap was unsafe. The warning used the Mira sound effect on an Android phone at maximum volume. A similar audio sound was used in our previous study of permissive alerts in which participants crossed the alerted gap 98% of the time. This result provides strong evidence the warning signal was clearly detectable.
The method to detect crossing-motion initiation used real-time position data from the tracking helmet worn by participants. The raw position data were first smoothed by applying a Gaussian blur filter on the five most recent positions. A trigger fired when the speed of the participant’s movement in the crossing direction was greater than a threshold value. To determine the threshold, we ran the method on 875 road-crossing trials from previous experiments. We selected a value such that the trigger detected the initiation of all 875 crossings. This conservative threshold biased avoidance of misses over the generation of false alarms.
When a crossing motion was detected, the warning system calculated the time to arrival of the next approaching vehicle. Based on previous experiments, the average time to cross the single lane is about 2.0 s. A 0.75-s buffer was added to this average crossing time to arrive at a safe crossing threshold value of 2.75 s. The cell phone sounded the auditory alarm whenever participants initiated a crossing motion and the time to collision to arrival of the tail vehicle was less than 2.75 s.
Procedure
Participants were fitted with a tracking helmet, shutter glasses, and a harness connected to a post at the back of the simulator to prevent them from walking into the front screen. Each trial began with the road clear of traffic. A continuous stream of vehicles approached from the left-hand side. Participants were told that they should wait until the lead vehicle passed (to prevent them from crossing in front of the stream of traffic) but then could wait as long as they wished before attempting to cross the road. Traffic generation ceased once participants reached the sidewalk on the other side of the road, allowing them to return to the starting position. Traffic was then again generated in the same fashion as described earlier.
Each participant performed three practice trials followed by 20 test trials. Participants in the control group crossed the road while holding an inactive cell phone. Participants in the texting groups were asked to respond to the texts they received as fast as they could. On the first practice trial, they crossed without texting; on the remaining two practice trials and 20 test trials, they received and responded to texts. Participants in the warning group were given a brief description of how the warning system worked between the second and third practice trials. They were instructed to hold the cell phone and lean toward but not cross the roadway when a vehicle was very close to the intersection. The cell phone detected the motion and warned the participant with an auditory alarm. Participants were given no explicit instructions about whether to cross the road based on the warnings. The experiment took approximately 30 min to complete. This research complied with the American Psychological Association Code of Ethics and was approved by the Institutional Review Board at the University of Iowa. Informed consent was obtained from each participant.
Data Recording and Performance Variables
The positions and orientations of all moving entities were recorded on every time step, including the participant’s head, the cell phone, and all vehicles. Text messages sent to and received from the participant were recorded along with the time step that the message was sent or received. Performance variables captured aspects of gap selection, movement timing, and gaze direction:
Waiting time: The sum of the temporal gap sizes seen prior to crossing.
Gap taken: The size of the gap crossed.
Timing of entry: The time between the participant and the rear of the lead car in the gap passed at the moment the participant entered the swath of the vehicles.
Road crossing time: The time from entering to clearing the swath of the vehicles.
Time to spare: The time between the participant and the front of the tail vehicle at the moment the participant cleared the swath of vehicles.
Collision: A road crossing was classified as a collision if the time to spare was ≤0.
Attention to traffic: Percentage of time the participant spent looking toward the traffic during a trial.
All measures were averaged across the 20 road-crossing trials to arrive at aggregate scores. We also computed a variability of timing of entry score.
Gaze Direction Estimation
A learning algorithm was used to develop a method to estimate participant gaze from head position and orientation in order to determine participants’ attention to the traffic, the cell phone, and elsewhere. Gaze classification was based on the position and orientation of the participant’s head relative to the cell phone and the vehicles on the road. Training and test data sets were collected in which the viewer’s gaze was known. The training data set served as the input to a support vector machine (SVM; Chang & Lin, 2011), which computed parameters for classification of gaze direction from data recorded during the experiment trials. At each moment of the simulation, the participant’s gaze was classified as directed either to traffic, to the cell phone, or elsewhere (“other”). The model returned by the SVM achieved 93% correct classification with the test data set.
Participants
The participants were 48 undergraduate students from an elementary psychology course who received course credit for their participation. There were 16 participants in each group, with eight females and eight males in both the control and texting groups, and nine females and seven males in the warning group.
Data Analytic Strategy
Mixed-effects logistic regression was used to model the likelihood of accepting (or rejecting) a gap based on condition and gap size. Log likelihood ratio tests indicated that the model fit best supported by the data included a random intercept for participant, a random slope for gap size, and fixed effects of gap size and condition. One-way analyses of variance (ANOVAs) were conducted for all other measures with condition (control, texting, warning) as a between-subjects factor. Fisher’s least square difference test was used for all post hoc tests with alpha = 0.05. There were no significant main effects or interactions involving gender; therefore, the analyses were collapsed over gender.
Results
Participants in the warning condition received a total of 164 warning signals on 318 road-crossing trials (two trials were dropped due to technical problems). Participants received a warning and entered the road on 28 trials. Surprisingly, participants never reversed their motion and returned to the side of the road on “warned” trials. The average gap size crossed on warned trials was 2.98 s. The average timing of entry (relative to the rear of the lead vehicle in the gap) on warned trials was 1.03 s, the average time to spare was 0.11 s, and there were collisions on 10 of the 28 trials. A warning was issued on every crossing that resulted in a collision for those in the warning condition (shown as 0.03 of all 328 crossings in Table 1). Of the remaining warnings, 107 were sent on trials on which the participant made a head movement but made no overt movement to cross the road. Another 29 warnings were sent on trials on which the participant made a head movement before the lead vehicle in the gap arrived at the intersection, which caused the warning to be sent. On 10 of these trials, the participant continued forward movement and crossed the next gap (just behind the lead vehicle that triggered the warning).
Means and Standard Deviations (in parentheses) of Performance Variables for Control, Texting, and Warning Conditions
Note. C = control; T = texting; W = warning.
Table 1 provides descriptive information for all performance variables for the control, texting, and warning groups and the results of group comparisons. These results include all of the warned trials.
Gap Selection
Waiting time
A significant main effect of condition, F(2, 45) = 4.08, p = .02, η2 = .15, indicated that participants in the warning condition waited significantly longer than those in the control and texting conditions, who did not differ from each other.
Mean gap size
A significant effect of condition, F(2, 45) = 3.24, p < .05, η2 = .13, indicated that participants in the texting condition accepted significantly smaller gaps than those in the warning condition. Participants in the control condition did not differ significantly from those in the warning or texting conditions.
Likelihood of taking a gap
Mixed-effects logistic regression analyses showed that participants in all conditions were more likely to choose larger rather than smaller gaps, z = 10.45, p < .001, with the average odds of accepting a gap increasing by 10.36 with each 0.5-s increase in gap size (Figure 4). In addition, gap choices in the texting group were significantly less conservative than those in the warning group, z = 2.12, p = .03, with 2.18 increased odds of accepting a given gap in the texting condition. There were no significant differences in gap choices between the control condition and either the warning or texting condition. Condition did not moderate gap selection sensitivity.
Logistic regression curves illustrating the likelihood of selecting gaps of different sizes for the control, texting, and warning conditions.
Movement Timing
Crossing time
There was no effect of condition for crossing time, F(2, 45) = .84, ns.
Timing of entry
A significant effect of condition, F(2, 45) = 5.80, p = .01, η2 = .21, indicated that participants in the warning condition timed their entry into the gap less tightly than those in the control condition but not the texting condition. The control and texting conditions did not differ significantly.
Variability of timing of entry
A significant effect of condition, F(2, 45) = 13.82, p < .001, η2 = .38, indicated that participants in the texting and warning conditions exhibited significantly more variability in their timing of entry than those in the control condition. The texting and warning conditions did not differ significantly.
Time to spare
There was no effect of condition for time to spare, F(2, 45) = .66, ns.
Collisions
There was no effect of condition for collisions, F(2, 45) = 2.68, ns.
Gaze Direction
Analysis of the mean percentage of time participants looked at the traffic revealed a main effect of condition, F(2, 45) = 37.75, p < .001, η2 = .63 (Figure 5). Each group differed significantly from the others, with those in the control condition spending the most time looking at traffic and those in the warning condition spending the least amount of time looking at traffic (Figure 5).
Estimation of percentage of time spent looking at the traffic, the phone, and elsewhere during the 2 s before and after initiation of road crossing for control (top), texting (middle), and warning (bottom) conditions.
Discussion
The overall goals of this experiment were to assess (a) the effects of texting on unsafe road-crossing behavior in pedestrians and (b) the effectiveness of warnings in reducing unsafe road crossing in texting pedestrians. As we expected, participants in the texting condition chose tighter gaps than those in the warning condition and spent less time looking at traffic than participants in the control condition. This is consistent with other research showing that pedestrians engage in riskier road-crossing behavior when distracted by texting (e.g., Hatfield & Murphy, 2007; Lin & Huang, 2017; Nasar & Troyer, 2013; Neider et al., 2010; Rahimian et al., 2016; Schwebel et al., 2012). We also found that warnings helped mitigate the harmful effects of texting on road-crossing behavior. Overall, participants in the warning group exhibited more cautious road-crossing behavior than the control and texting groups, waiting significantly longer and choosing larger gaps for crossing. However, unlike participants in our previous “permissive alerts” study who almost always complied with alerts indicating safe gaps, participants in the current “prohibitive alerts” study never heeded warnings about unsafe gaps once they entered the roadway. Finally, contrary to our expectations, we found that participants in the warning group spent less time looking at the traffic than those in the control or texting groups (though the differences were smaller than those in Rahimian et al., 2016). It is unclear why participants in the warning condition spent less time looking at traffic, because the warning provides no assistance in selecting a gap.
An important question these results raise is why participants in the warning condition were more cautious in their road crossing. One explanation is that the warnings came sufficiently early in the decision-making process that they inhibited participants from crossing smaller gaps. Another possible explanation is that they found the warnings aversive and hence tried to avoid triggering the warning by choosing larger gaps. A third explanation is that warnings heightened participants’ awareness that their gap choices were being monitored, and therefore they were more attentive to choosing safe gaps than were the participants in the other conditions. Quite likely, all three of these factors contributed to greater caution in the warning group.
The most surprising finding was that participants in the warning group never aborted a crossing once they made a deliberate crossing movement. They received a warning on all crossings that resulted in a collision. Moreover, the warnings were highly predictive of risk—they had collisions on over a third of the trials and near misses on the rest of the trials on which they received a warning and crossed the road. This is consistent with a recent study using a pedestrian simulator similar to ours that examined vibrotactile alerts sent to nontexting pedestrians when an approaching gap was dangerously small (Cœugnet et al., 2017). Similar to our experiment, participants made safer crossings overall but crossed almost 50% of the alerted gaps.
Why do participants ignore warnings? One possibility is that the perceived risk in a simulated road-crossing task is insufficient to cause them to alter their behavior. Another possibility is that they judge that it is quicker and safer to finish crossing than to reverse direction and return to the curb. Prior research on response inhibition has shown that once an action is initiated, changes to the action (including stopping the action) are suppressed (Verbruggen & Logan, 2008). The suppression of alternative actions is particularly strong when the alternative requires reprogramming the action (Gray, 2009). This finding may also help explain the difference in how participants responded to the permissive alerts in our earlier study versus the prohibitive warnings in the current study. In particular, the permissive alert was sent approximately 1 s before the gap arrived at the crossing line, whereas the prohibitive warning was sent only once participants made a forward movement. As a result, participants had far more time to regulate their actions in the permissive-alert than in the prohibitive-warning condition. For warnings to be effective, it may be necessary to send warnings before a crossing action is initiated. However, it may be difficult to detect an intention to cross sufficiently early to inhibit the crossing action, particularly in real roadway environments. Moreover, a highly sensitive threshold for detecting crossings likely increases the frequency of warnings and may lead users to interpret them as false alarms, which can lead to reduced trust and compliance (Lees & Lee, 2007).
Although these results offer promise for the use of mobile communications technology in promoting safe road crossing, the degree to which participants in the warning group focused on the cell phone raises concerns about overreliance on technology for guiding road crossing. The reduced attention to traffic could leave them vulnerable to unexpected changes in traffic or technological failures in predicting gap affordances, resulting in unsafe entry into traffic-filled roadways. Extensive testing of such assistive technologies is critical before taking steps to deploy them on real roads. Simulation can be a particularly useful tool for safely and systematically testing these systems. Among the advantages of using a pedestrian simulator are that the motions of vehicles are precisely controlled and the movements of participants can be accurately captured. This design makes it possible to study the effectiveness of assistive technologies under optimal conditions. However, even under these ideal conditions, we found that warnings were ineffective in preventing participants from crossing dangerously small gaps once they began to cross.
Care is warranted in generalizing the results beyond the specific conditions examined in this experiment. Both the repeated crossings of a dense stream of traffic and the awareness that they were being observed likely primed participants’ attention. Although the total of 5,376 U.S. pedestrian fatalities in 2015 is alarmingly high, the likelihood that an individual pedestrian will be involved in a crash on any particular road crossing is very low. Pedestrian assistive technology must aim to address these tragic but rare and unexpected events by providing useful information at the right time and place. Real-time information about when roads are safe or dangerous to cross could aid pedestrians in making good crossing decisions. However, there are significant challenges in the development of sensor technology to reliably and accurately measure traffic conditions and movement initiation and in the design of effective interfaces to provide information in a useful form when needed.
In conclusion, this research highlights the adverse effects of pedestrian texting on safe road crossing and both the potential and challenges of using cell phone warnings to reduce the risk of vehicle–pedestrian collisions when walking across roads while texting. The ideal solution is to discourage pedestrians from texting while crossing roads. Several communities have recently made it illegal to text and walk. However, recognizing that many pedestrians will continue to use mobile devices, it is important to explore how V2P technologies can assist pedestrians. This experiment demonstrates the effectiveness of warnings to improve road-crossing decisions of texting pedestrians. However, the reduction in attention to traffic for those who received warnings is a concern that requires additional investigation. Additional research is also required to understand why participants did not respond to warning received once they initiated a crossing. In future work, we plan to examine these factors and also how alerts and warnings can aid nontexting pedestrians.
Key Points
Mobile-device use puts pedestrians at greater risk for collisions with vehicles.
An immersive pedestrian simulator was used to test how texting pedestrians respond to warnings about traffic conditions sent to a cell phone.
Texting participants who received warnings waited longer to cross than either the control group or the texting-only group. They also chose larger gaps than those in the texting-only group.
Texting participants who received warnings spent less time looking toward traffic than those in the texting-only group and the control group (who spent the most time looking at traffic). This behavior may lead to reduced situation awareness.
Texting participants who received warnings failed to abort unsafe crossings.
Mobile-communications technology offers promise for increasing road safety, but more study of how pedestrians respond to warnings is needed to assess its effectiveness.
Footnotes
Authors’ Note
This research was supported by National Science Foundation Awards BCS-1251694 and CNS-1305131 and by the U.S. Department of Transportation, Research and Innovative Technology Administration, Prime DFDA No. 20.701, Award No. DTRT13-G-UTC53, through the SAFER-SIM University Transportation Center.
Pooya Rahimian is a PhD candidate at the University of Iowa in the Computer Science Program. He received an MS degree from East Carolina University in software engineering in 2013.
Elizabeth E. O’Neal is a PhD candidate in the Department of Psychological and Brain Sciences at the University of Iowa. She received an MPH from the University of Iowa in 2016.
Shiwen Zhou is an undergraduate in the Department of Psychological and Brain Sciences at the University of Iowa. She received a high school diploma from Hefei No. 6 Senior High School (China).
Jodie M. Plumert is a professor at the University of Iowa in the Department of Psychological and Brain Sciences. She received a PhD in child psychology from the University of Minnesota in 1990.
Joseph K. Kearney is a professor at the University of Iowa in the Department of Computer Science. He received a PhD in computer science from the University of Minnesota in 1983.