This study explores the potential of virtual reality (VR) to enhance construction safety training by eliciting critical emotional responses, particularly in novices lacking firsthand experience with hazardous scenarios. By targeting emotions like fear and distress, VR aims to improve hazard recognition and foster risk-averse behaviors, addressing gaps in traditional training methods.
The study involved 55 construction management students using a VR simulation integrated with haptic feedback to replicate accident scenarios. Emotional responses were measured through galvanic skin response (GSR) metrics, pre- and post-experience self-reported questionnaires and semi-structured interviews. Participants navigated a virtual construction site, identified hazards and experienced a simulated accident to assess emotional and behavioral impacts.
The VR simulation significantly increased emotional arousal, with GSR data showing a strong effect size and significant increases in distress, guilt, fear and shame. Participants reported heightened empathy, guilt and responsibility during post-VR interviews. These findings suggest that VR effectively engages users emotionally, bridging the experiential gap in traditional training by replicating hazardous scenarios with high realism and impact.
The study highlights VR's potential as a scalable, immersive training tool for eliciting emotional and physiological precursors associated with risk-averse decision-making in the construction industry. By evoking anticipatory emotions theorized to influence risk perception, VR may help prepare novices to engage more cautiously with hazardous scenarios, though the link between these precursors and actual safety behavior requires field validation.
This research provides empirical evidence of VR's capability to elicit emotional responses that influence decision-making in hazardous environments. It offers valuable insights for developing innovative, emotionally engaging safety training programs for the construction sector.
1. Introduction
1.1 Motivation
The construction industry is fraught with complex and hazardous environments requiring workers to make critical safety decisions. Effective hazard recognition and management are essential to reduce risks and ensure worker safety. Despite various training protocols, the industry's accident rates remain high, highlighting the need for more effective training approaches. This study investigates how virtual reality (VR) can enhance safety training by eliciting emotional responses in construction novices, potentially improving their hazard recognition and decision-making skills. It has been shown that hazard recognition performance on construction sites remains poor, leading to unsettling statistics on accidents and fatalities in the construction industry (Ichniowski, 2020; Wiatrowski and Janocha, 2014). To address this issue, the research community has focused on developing training protocols and interventions to improve hazard recognition skills among practitioners. However, a more effective approach to improving hazard recognition skills should not only focus on the ability to identify hazards but also consider the instances where workers may choose not to report a hazard they have identified. Various reasons, such as social pressure, high risk tolerance and propensity for risk-taking behavior can influence this decision (Bhandari and Hallowell, 2022; Salas et al., 2020). Therefore, while providing training for hazard recognition may enable workers to identify a hazard during construction, if the worker decides not to report and consequently manage that hazard, the risk of an accident continues to loom over. Training must focus on more than hazard recognition. Decision-making under risky conditions, often influenced by social pressure or risk tolerance, is equally critical.
It is known that emotions, particularly anticipatory emotions such as fear, anxiety and apprehension, play a key role in decision-making under risky situations (Lerner and Keltner, 2000; Loewenstein et al., 2001). As it relates to the construction domain, this implies that the emotions that emerge in anticipation of an accident, play a role in how risky the hazard is perceived by a worker and, consequently, how likely they are to report the hazard (Wang et al., 2021a, b; Zohar, 2000). Interestingly, it was found that the intensity of anticipatory emotions that drive risk-aversion is dependent on the person's ability to vividly imagine the unwanted outcome. While this ability to create vivid mental imagery is difficult to measure and predict, it can be said that past experiences with an accident make it easier to imagine its possibility in the future (Corgnet et al., 2020; Miller et al., 1987; Villejoubert and Vallée-Tourangeau, 2010; Zaleskiewicz and Traczyk, 2020). This is in line with the popular availability heuristic, which suggests that people tend to estimate the likelihood of an event based on the ease with which relevant examples come to mind (Tversky and Kahneman, 1974). Novices without experience lack past experience of witnessing accidents on construction sites. Therefore, there is an opportunity for training programs to provide enhanced learning experiences that not only provide background on fundamental hazard identification concepts, but also target crucial emotions to drive risk aversion in practice. Anticipatory emotions like fear play a critical role in decision-making under risk, and VR, by simulating accidents, may effectively evoke these emotional precursors that are theorized to influence risk-averse decision-making.
Virtual reality (VR) provides compelling medium for this type of learning particularly, for high-risk scenarios due to its capability to simulate accident experiences for novices without exposing them to real dangers (Albert et al., 2014a; Bailenson, 2018; Eiris et al., 2020b). In addition to visual and auditory stimuli, VR can engage various senses, such as touch, through the use of haptic feedback suits. This multisensory immersion can enhance the realism and emotional impact of the VR experience (Ahmed et al., 2016; Basdogan et al., 2001; Menelas and Benaoudia, 2017; Patil et al., 2018). However, there is a lack of empirical evidence on the effectiveness of this approach in influencing construction novices’ emotional responses, which could impact their subsequent risk-related behaviors. As a result, there is a pressing need to assess whether VR can help bridge this experiential gap for construction novices, who may lack firsthand exposure to accidents on construction sites, by providing an emotionally arousing virtual experience (Guo et al., 2012; Mastli and Zhang, 2017).
In this study, the authors aim to investigate the potential of VR, used in conjunction with haptic feedback, in eliciting emotional reactions and investigating its effect on construction novices’ perspectives on safety when confronted with risky situations in the construction domain. The VR experience was developed using findings grounded in previous research by the authors to simulate accident scenarios for novices in a realistic setting. To assess the impact of the VR experience on users’ emotional arousal and risk-tolerance, the authors administered pre- and post-experience questionnaires and collected physiological data in the form of galvanic skin response metrics (GSR) that measure participants’ responses when presented with accident scenarios. Our hypothesis is that the VR experience will have a significant impact on users’ emotional arousal, subsequently impacting their risk tolerance on the construction site.
1.2 External validity: CM students vs. practitioners
Our sample comprises upper-division construction management (CM) students, who can be considered novices. Novices typically possess fewer and less richly indexed hazard schemas, have fewer vivid accident exemplars and exhibit less entrenched reporting norms. These characteristics may increase the instructional value of emotionally vivid simulations. Practitioners, by contrast, bring deeper experiential priors, stronger production pressures and social-norm trade-offs that can change how they respond to training.
We therefore interpret our findings as effects on novice arousal and safety-relevant appraisals. To support ecological validity, scenario content and pacing were vetted in consultation with construction professionals (see section 4.3.1). A prospective replication with a practitioner sample is outlined in the Future Work section.
1.3 How VR enhances safety training: proposed mechanisms
We propose four complementary mechanisms through which virtual reality can enhance safety training: (1) ecological fidelity and presence increase the realism and recognizability of safety-critical cues; (2) anticipatory emotions such as fear, distress and guilt reduce risk tolerance and increase willingness to intervene; (3) multisensory salience (including haptic feedback) strengthens physiological arousal and memory encoding and (4) active interaction (for example, navigation, capture and reporting affordances) supports consolidation of cue–action associations.
In this study, event-related galvanic skin response (GSR), pre- to post-scene self-reported emotions, spontaneous protective behaviors observed in scene and interview content serve as indicators of internal states and proximal actions that are theorized precursors to safer choices.
For the decision-making framework, we conceptualize safety decision-making in hazardous contexts as a sequence from (1) arousal and appraisal (initial physiological/emotional response and evaluation of danger) to (2) intention to act (willingness to report/intervene) to (3) proximal action (immediate protective responses). In our data, arousal is indexed by event-related GSR and pre/post emotion ratings; appraisal is not measured directly and is instead inferred qualitatively from interview language about danger/realism considered alongside arousal; intention is inferred from interview statements about responsibility/willingness to act; and proximal action is indexed by spontaneous in-scene protective behaviors (e.g. calling out, reaching). We do not claim downstream behavior change beyond the session; that link is a target for future work.
2. Background
Research highlights a critical gap in workers’ preparedness to identify and manage risks on job sites (Albert et al., 2014a; Perlman et al., 2014). Despite concerted efforts to improve hazard recognition through training, the expected outcomes have not been consistently realized across construction sites (Namian et al., 2016; Zuluaga et al., 2016). This discrepancy can be attributed to training modules that lack external and ecological validity, fail to transfer knowledge effectively or do not engage adult learners in meaningful learning contexts (Albert et al., 2014a; Bhandari et al., 2019; Jeelani et al., 2020; Demirkesen and Arditi, 2015).
2.1 Lack of ecological validity in construction safety training and research
Traditionally, researchers have used photographs and two-dimensional videos to measure and teach safety skills such as hazard recognition and risk assessment (Uddin et al., 2020; Zuluaga et al., 2016; Eiris et al., 2021). However, the need for a more realistic presentation of stimuli is evident from the methodological limitations of previous studies, which relied on pictures and videos of construction sites for assessing hazard recognition performance. These methods, while contributing to the development of training programs, fall short in predicting performance effects on real construction sites due to the simplicity of the assessment environment. Furthermore, studies utilizing pictures to observe construction workers’ eye movements and cognitive load have added significantly to the understanding of worker behavior, yet they underscore the difference in perception and cognitive processes when viewing planar versus three-dimensional stimuli, as found on actual construction sites (Gibson, 2014; Reichelt et al., 2010; Todd, 2004).
2.2 The role of virtual reality in construction safety training
Recognizing these limitations, recent efforts have turned towards VR as a means to provide a more immersive and effective learning experience (Eiris et al., 2018; Jeelani et al., 2020; Pham et al., 2018; Zhang et al., 2020; Kim, 2022; Babalola et al., 2023). VR's ability to simulate real-life scenarios in a controlled environment offers a unique advantage over traditional lecture-based training, potentially improving hazard recognition and emotional arousal, which are crucial for effective learning and long-term retention (Aggarwal et al., 2006; Farra et al., 2013; Mastli and Zhang, 2017; Slater, 2003).
A comprehensive review by Moore and Gheisari (2019) outlined how VR and MR technologies have been employed over the last decade for safety-related applications within construction, including training, hazard monitoring and preconstruction planning. The review detailed the incorporation of various layers of information, such as building information modeling (BIM), real-time geographical location, and audio alerts, to create immersive and information-rich experiences. These technological interventions aim to transfer knowledge more effectively to workers, actively warn them of site hazards and eliminate hazards during the preconstruction phase. Another study focusing on the efficiency of VR-based safety training, particularly in the context of heavy machinery operations, highlighted the significance of immersive VR training over traditional flat screen displays. The research revealed that VR headsets offer a more immersive experience, providing greater realism and depth perception, which in turn enhances the accuracy of hazard identification for critical hazards, such as electric cables.
2.3 Emotional arousal and learning in VR
Recent literature underscores a more nuanced understanding of virtual reality (VR) in construction safety training, particularly emphasizing the importance of risk perception and training satisfaction. A study examined the effectiveness of VR safety training among 248 construction workers (Yoo et al., 2023). It highlighted how trainees’ risk perception significantly influences their satisfaction with VR safety training, which in turn affects its overall effectiveness. The research suggests that while VR technology is advancing in safety training applications, understanding the psychological and technological aspects is crucial for improving training outcomes. This approach sheds light on the importance of considering both the technological capabilities of VR and the psychological processes of trainees to enhance safety training's effectiveness.
Emotional engagement is a precursor to improved learning outcomes, as it enhances neuroplasticity and facilitates intrinsic motivation towards learning (Bradley et al., 1992; Cahill and McGaugh, 1998; Goetz et al., 2007). Studies across various contexts, including safety training, have demonstrated that VR can sustain targeted emotional arousal effectively, thus promoting a more engaging learning environment (Wang et al., 2018).
VR's capacity to simulate realistic work environments and engage users in interactive hazard recognition exercises presents a significant advancement over traditional safety training methods. By facilitating direct interaction with hazards and potential consequences in a risk-free setting, VR training aligns with the cognitive and emotional needs of adult learners, offering a more personalized and impactful learning experience (Demirkesen and Arditi, 2015; Hannum, 2009; Albert et al., 2014a).
2.4 Knowledge gap
Despite the advancements in hazard recognition training, training efforts have not effectively targeted emotional factors that impact the risk tolerance among workers, even though emotions associated with previous experiences of accidents play a crucial role in decision-making processes. This is especially critical for novice workers who may not yet have prior experience to draw upon for safety decision-making. VR may provide value for providing an immersive, yet safe, environment for this kind of learning, but there is a lack of empirical evidence on the effectiveness of VR in influencing construction novices’ emotional responses. A systematic investigation of the potential benefits of VR in this context could provide valuable insights for the development of innovative training solutions that address both hazard recognition and risk tolerance. This study seeks to examine the impact of emotionally engaging and realistic VR experiences on construction novices' emotional arousal. By addressing this gap, the present research aims to contribute to a more comprehensive understanding of the role that immersive technologies like VR can play in making construction sites safer.
2.4.1 Hypothesis
Prior work links anticipatory emotions (e.g. fear, distress) to lower risk tolerance and safer choices under hazard (Lerner and Keltner, 2000; Loewenstein et al., 2001). In immersive contexts, heightened arousal supports memory encoding and salience of safety-critical cues (Cahill and McGaugh, 1998), while presence may elicit spontaneous protective responses. Accordingly, we hypothesized increased arousal (H1), increased anticipatory emotions (H2) and exploratory in-scene protective behaviors (H3).
(Arousal): The accident event will elicit a significant increase in physiological arousal, indexed by event-related GSR during the accident window relative to a pre-event period in the same session.
(Emotions): Self-reported anticipatory emotions (fear, distress, guilt) will increase from pre-scene to post-scene.
(Exploratory): The simulation will elicit spontaneous protective behaviors in scene that are consistent with heightened intent to intervene.
3. Methodology
3.1 Overview
To address the knowledge gap defined above, a mixed-methods subject test was conducted related to a hazard identification scenario that included a simulated accident aimed at targeting emotions. In order to understand the impact that this mode of learning had on users’ emotions, the researchers collected pre- and post-activity survey responses, GSR data to analyze the impact of the experience on participants emotional arousal and risk-perception, and semi-structured interviews to gain further insight into participants’ experience and its impact. The detailed procedures followed for these methods are described in subsequent sections.
3.2 Participants
Participants for the study were recruited from a course in the construction management program at Arizona State University (ASU). Convenience sampling was utilized to recruit participants from a sample of students in the construction field who were in their third or fourth year of study. These students were selected as they were committed to pursuing construction careers but generally had limited practical experience in the industry, making them ideal candidates for studying the impact of the developed VR experience on construction novices. Participants received course extra credit as an incentive for participation. To receive the credit, students were required to complete the VR activity and submit a written report on hazard recognition. Participation was entirely voluntary, with alternative opportunities for equivalent extra credit available to those who chose not to participate. By recruiting primarily third and fourth-year students who were dedicated to pursuing careers in construction, the authors aimed to capture a group of participants who were knowledgeable about the field but lacked extensive hands-on experience. This ensured that the study focused on novices who were at a critical stage in their training and were actively building their hazard recognition and risk tolerance skills.
3.3 The virtual reality experience
3.3.1 Development
The simulation ran on a PC-based virtual reality system (HTC Vive Pro head-mounted display using the SteamVR runtime) with haptic feedback delivered via a bHaptics vest that produced physical feedback time-locked to impact events. The virtual site was authored in Unity using physics-based interactions, SketchUp/Blender for modeling and Mixamo-rigged avatars with animations captured in Rokoko Studio. Event markers were written at the onset of the accident to synchronize the VR timeline with GSR and observer notes. Assets were reusable across sessions to support consistent delivery (see Figure 1, participants wore the VR headset, VR controllers and haptic suit as shown). The virtual environment illustrated in Figure 2 was developed based on findings from previous studies: one study outlined design considerations for simulating hazardous scenarios.
The photograph features a person standing in an indoor setting, wearing and holding several pieces of specialized technology that are identified by text labels and arrows. On the head, the person wears a black device labeled “H T C V I V E V R Headset” with an arrow pointing toward it. The person holds a black device in each hand, which is labeled at the bottom left as “V R Controllers” with an arrow pointing toward the equipment. The torso and arms are covered by a black vest and arm sleeves, labeled on the right as “haptic feedback suit” with two arrows pointing toward the chest and the left forearm.Participant geared with VR headset and haptic suit. Source: Authors’ own work
The photograph features a person standing in an indoor setting, wearing and holding several pieces of specialized technology that are identified by text labels and arrows. On the head, the person wears a black device labeled “H T C V I V E V R Headset” with an arrow pointing toward it. The person holds a black device in each hand, which is labeled at the bottom left as “V R Controllers” with an arrow pointing toward the equipment. The torso and arms are covered by a black vest and arm sleeves, labeled on the right as “haptic feedback suit” with two arrows pointing toward the chest and the left forearm.Participant geared with VR headset and haptic suit. Source: Authors’ own work
The digital illustration features a desert landscape where a pipeline project is underway, with various pieces of heavy machinery and several workers present. In the foreground on the left, a large rock sits near a pickup truck, and three workers stand nearby. Along the center, a long pipe extends into the distance, passing through several white portable structures. Multiple large yellow excavators with crane attachments are positioned along the pipeline to assist with the task. Additional workers are visible near the portable structures on the right side of the pipe. The background consists of rolling sand hills and sparse desert vegetation under an overcast sky.Virtual reality environment developed based on previous findings. Source: Authors’ own work
The digital illustration features a desert landscape where a pipeline project is underway, with various pieces of heavy machinery and several workers present. In the foreground on the left, a large rock sits near a pickup truck, and three workers stand nearby. Along the center, a long pipe extends into the distance, passing through several white portable structures. Multiple large yellow excavators with crane attachments are positioned along the pipeline to assist with the task. Additional workers are visible near the portable structures on the right side of the pipe. The background consists of rolling sand hills and sparse desert vegetation under an overcast sky.Virtual reality environment developed based on previous findings. Source: Authors’ own work
VR Patil et al. (2023b), while another identified key aspects of a virtual construction environment that were crucial for construction practitioners to experience a strong sense of presence Patil et al. (2023a). Based on these, realistic virtual construction co-workers were developed for the VR experience that were found to facilitate emotional engagement and presence. To enhance the realism of the virtual construction workers, avatars used in the VR simulation were generated by recording realistic human movements through the Rokoko motion capture suit and applying those animations to realistic 3D models of construction workers.
To ensure the authenticity and relevance of the accidents modeled in the VR simulation, industry focus groups were involved in the validation process. These focus groups comprised experienced professionals from the construction industry who provided feedback and insights on the realism and accuracy of simulated processes and accidents. Their input was invaluable in ensuring that the accidents depicted in the VR experience reflected real-world construction hazards and scenarios.
3.3.2 The participant experience in VR
The VR simulation consisted of two distinct phases. The initial phase served as a preparatory training ground where participants were acclimated to the technical aspects of the VR system. During this phase, they were tutored on how to navigate the virtual space using the VR controllers, how to interact with objects within the environment and other technical maneuvers essential for a smooth user experience. Importantly, this training environment was designed to be context-neutral, devoid of any construction-specific elements or scenarios to avoid any chance of inadvertently priming participants or affect their subsequent performance in recognizing construction-specific hazards.
Once the participants were comfortable with using the VR equipment, they progressed to the virtual construction site developed for this study. Participants were tasked with navigating a virtual construction site where they were instructed to conduct a hazard recognition activity and document any hazards that they noticed. Participants had the ability to report the hazards that they noticed by communicating with the research assistant and also by capturing a photograph using a virtual iPad, which was a part of the VR simulation. After participants declared that they had identified all hazards, they were asked to hand over the iPad to the site supervisor avatar. Meanwhile, the research assistant discreetly used a keyboard input to position the supervisor in such a way that, when the participant faced him to hand over the iPad, the simulated accident would occur directly behind them and clearly within the participant's field of view (See Figure 3 for the participant's point of view). Once the participant was in position, the accident was triggered and simulated on the virtual construction site. During the VR experience, participants wore haptic feedback suits that were designed to enhance the realism of the experience. These devices were integrated into the VR setup and provided tactile sensations to the participants. Specifically, when the accident was simulated behind the site supervisor avatar, the haptic devices worn by the participants were triggered and started vibrating, providing a physical sensation synchronized with the visual and auditory cues of the accident (see Figure 3).
The illustration on the left features a worker in the foreground and another worker standing beneath a suspended portable structure, with a circle around the second worker and a vertical arrow pointing down to the label “Gravity Hazard”. The illustration on the right is a grayscale version of the same scene, but it depicts the suspended structure falling onto the worker, with a circle around the impact area and a curved arrow pointing down to the label “Accident”. In both illustrations, a large excavator with a crane attachment is positioned next to the portable structure.POV of participant. (left) Hazard depicting a virtual worker standing under the shack assembly. (right) Simulation of an accident where the shack assembly falls onto the virtual construction worker. Source: Authors’ own work
The illustration on the left features a worker in the foreground and another worker standing beneath a suspended portable structure, with a circle around the second worker and a vertical arrow pointing down to the label “Gravity Hazard”. The illustration on the right is a grayscale version of the same scene, but it depicts the suspended structure falling onto the worker, with a circle around the impact area and a curved arrow pointing down to the label “Accident”. In both illustrations, a large excavator with a crane attachment is positioned next to the portable structure.POV of participant. (left) Hazard depicting a virtual worker standing under the shack assembly. (right) Simulation of an accident where the shack assembly falls onto the virtual construction worker. Source: Authors’ own work
After the accident, the simulation paused and went into a black-and-white mode to signify a pause from the VR experience to have a conversation with the research assistant about what had just happened. At this point, the participants were asked what they had witnessed and their reaction to the accident. This audio was recorded and converted into transcripts for data analysis.
3.4 Research protocol
The following sections describe each of the steps of the research protocol, which are illustrated in Figure 4.
The flowchart consists of six primary rectangular boxes at the center, arranged horizontally from left to right. A rectangular box to the center left labeled “Pre-V R survey” contains two rounded rectangular boxes below labeled “Emotions” and “Risk-perception” and connects above to a dashed rounded rectangular box labeled “survey responses”. A horizontal arrow from the “Pre-V R survey” box connects to a rectangular box on the right labeled “Gearing participant”, which contains three rounded rectangular boxes below labeled “V R headset”, “Haptic suit”, and “G S R sensor”. A horizontal arrow from the “Gearing participant” box connects to a rectangular box on the right labeled “V R Training”, which contains two rounded rectangular boxes below labeled “Navigation” and “Documentation”. A horizontal arrow from the “V R Training” box connects to a rectangular box on the right labeled “Hazard Recognition Activity”, which contains three rounded rectangular boxes below labeled “Recognition”, “Accident Simulation”, and “Haptic feedback” and connects above to two dashed rounded rectangular boxes labeled “Participant audio” and “G S R data”. A horizontal arrow from the “Hazard Recognition Activity” box connects to a rectangular box on the right labeled “Post-V R survey”, which contains three rounded rectangular boxes below labeled “Emotions”, “Risk-perception”, and “Demographic data”, and connects above to a dashed, rounded rectangular box labeled “survey responses”. A horizontal arrow from the “Post-V R survey” box connects to the final rectangular box on the right labeled “Post-V R Interview”, which contains a rounded rectangular box below labeled “Semi-structured” and connects above to a dashed rounded rectangular box labeled “Interview transcripts”. A legend at the bottom left defines the shapes: a rectangular box equals “Protocol Step”, a rounded rectangular box equals “Sub-step”, and a dashed rounded rectangular box equals “data collected”.Research protocol. Source: Authors’ own work
The flowchart consists of six primary rectangular boxes at the center, arranged horizontally from left to right. A rectangular box to the center left labeled “Pre-V R survey” contains two rounded rectangular boxes below labeled “Emotions” and “Risk-perception” and connects above to a dashed rounded rectangular box labeled “survey responses”. A horizontal arrow from the “Pre-V R survey” box connects to a rectangular box on the right labeled “Gearing participant”, which contains three rounded rectangular boxes below labeled “V R headset”, “Haptic suit”, and “G S R sensor”. A horizontal arrow from the “Gearing participant” box connects to a rectangular box on the right labeled “V R Training”, which contains two rounded rectangular boxes below labeled “Navigation” and “Documentation”. A horizontal arrow from the “V R Training” box connects to a rectangular box on the right labeled “Hazard Recognition Activity”, which contains three rounded rectangular boxes below labeled “Recognition”, “Accident Simulation”, and “Haptic feedback” and connects above to two dashed rounded rectangular boxes labeled “Participant audio” and “G S R data”. A horizontal arrow from the “Hazard Recognition Activity” box connects to a rectangular box on the right labeled “Post-V R survey”, which contains three rounded rectangular boxes below labeled “Emotions”, “Risk-perception”, and “Demographic data”, and connects above to a dashed, rounded rectangular box labeled “survey responses”. A horizontal arrow from the “Post-V R survey” box connects to the final rectangular box on the right labeled “Post-V R Interview”, which contains a rounded rectangular box below labeled “Semi-structured” and connects above to a dashed rounded rectangular box labeled “Interview transcripts”. A legend at the bottom left defines the shapes: a rectangular box equals “Protocol Step”, a rounded rectangular box equals “Sub-step”, and a dashed rounded rectangular box equals “data collected”.Research protocol. Source: Authors’ own work
3.4.1 Pre-VR survey
Before entering the VR experience, participants were asked to complete a baseline survey to assess their current emotional state. The survey used was a modified version of the Post-Film Questionnaire developed by Rottenberg et al. (2007), which has been validated in various studies across disciplines. Participants were presented with a list of emotions and were asked to rate the extent to which they were experiencing each emotion at that moment on a scale from 0 (not at all) to 8 (extremely). The emotions included both positive and negative affective states such as happiness, sadness, fear, distress, guilt, shame, enthusiasm and determination, among others relevant to the context of the study. Participants were also provided with the opportunity to write in any additional emotions they felt that were not included in the provided list. This approach allowed for a comprehensive assessment of the participants' emotional states prior to the VR experience, ensuring that any changes observed post-experience could be attributed to the VR simulation.
3.4.2 Gearing participant
In this phase, participants were equipped with a haptic suit, VR headset and GSR sensor sequentially. For the GSR measurements, we utilized a Shimmer3 GSR + Unit, which is a sensor designed for capturing electrodermal activity, providing a reliable measure of physiological arousal Boucsein (2012). The GSR sensor comprised two Ag/AgCl electrodes attached to the palmar surfaces of the participants' left hand fingers – specifically, the middle and ring fingers – using velcro straps. The selection of these fingers is standard practice due to their high density of eccrine sweat glands, which enhances the sensitivity of the measurements Dawson et al. (2017).
The electrodes were connected to the Shimmer3 device, which participants wore like a wristband on their left wrist. This setup allowed for continuous recording of skin conductance data throughout the VR experience without restricting the participants’ movements or interfering with the VR equipment. The GSR data was recorded and stored locally on the device's SD card for subsequent analysis.
3.4.3 VR training
After gearing up with the required equipment, participants were first put in a training environment before entering the virtual construction site. In the training environment, participants were instructed on how to use the controllers to be able to navigate inside the virtual environment. Participants were also trained on how to access and click pictures on the iPad that was made available to them in the VR simulation. The participants were asked to navigate the training environment and practice taking pictures on the iPad to confirm their familiarity with the VR equipment before moving them on to the virtual construction site, which was developed for this study to conduct hazard recognition activity.
Although the training environment was designed to be context-neutral, participants may have anticipated hazard-related scenarios once they transitioned into the main VR simulation. This could have introduced a degree of priming, as participants were aware that the purpose of the study was related to hazard recognition. To minimize this bias, no explicit mentions of accidents or hazardous situations were made during the initial training phase. Furthermore, participants were only instructed to explore and report hazards in the final phase, and no further interventions or cues were provided to suggest that an accident would occur.
3.4.4 Hazard recognition activity
Upon entering the virtual construction site, participants were informed that the hazard recognition activity had begun, and they were instructed to navigate the site to report and document any hazards they identify (see section “Participant Experience in VR” for more details). No further instructions were given after this point, and participants were given time to explore the construction site and locate hazards at their own pace until they mentioned they had completed the hazard recognition activity. Reporting and photo capture were available as behavioral traces but were not analyzed in the present study. A typical participant took approximately 30–40 minutes to complete the entire VR experience, which included the training, the hazard recognition task and the simulation of a construction accident.
3.4.5 Post-VR survey
Immediately after the VR experience, participants completed the same questionnaire as the pre-VR survey to assess any changes in their emotional states. By comparing the pre- and post-VR survey responses, we aimed to identify specific emotions that were influenced by the VR simulation. This method provided both quantitative and qualitative data on the emotional impact of the VR experience.
3.4.6 Post-VR interview
Next, participants took part in a semi-structured interview, which was included to gain insight into the impact that the VR experience had on their emotions and perspective on safety training. The following guiding questions were used to structure the exploratory nature of the interview:
How did you feel in the virtual environment?
How did you feel when the accident took place?
To what level were you able to empathize with the workers involved in the accident scene?
Were there any elements of the VR experience that you found particularly realistic or unrealistic?
These open-ended questions allowed an exploration of the participants' emotions and their overall qualitative reaction to the VR experience and its impact on their perception of safety on construction sites.
3.5 Procedure timeline
The study followed a standardized procedure. Approximate elapsed times are reported relative to the onset of the simulated accident (T = 0).
T – 20 min: Participants provided informed consent, completed the pre-VR survey and had GSR electrodes applied.
T – 10 min: Participants were equipped with the haptic vest and VR headset; calibration was performed.
T – 8 min: Neutral VR familiarization and training (navigation and photo capture practice).
T = 0: Participants entered the construction site and performed hazard recognition.
T + X: Accident event triggered; GSR event marker recorded; research assistants noted any spontaneous protective behaviors.
T + 2 min: Simulation paused; participants responded to a brief in-VR debrief prompt.
T + 5 min: Participants removed the VR equipment and completed a post-scene questionnaire.
T + 10 min: Semi-structured interview conducted (covering emotions, realism, responsibility and willingness to act)
4. Data analysis
To examine how VR experiences may influence safety-related responses, we first mapped the constructs measured in this study against a theoretical decision-making framework. While the full decision process in hazardous contexts likely involves multiple stages, our analysis focused on specific components that could be captured with the available methods.
We directly measured (1) affective and physiological responses, through event-related changes in galvanic skin response (GSR) during the accident window and pre/post changes in self-reported fear, distress and guilt and (2) proximal action indicators, through spontaneous protective behaviors observed during the accident (e.g. warning vocalizations, reaching gestures). We assessed indirectly, through interviews, (1) appraisal indicators, reflected in participants’ descriptions of the scene as “dangerous” or “realistic,” interpreted alongside physiological and emotional data and (2) stated intentions, expressed as willingness to report or intervene.
Without validated risk-perception scales or behavioral follow-up, we cannot directly measure cognitive appraisal or confirm whether stated intentions translate to actual safety behaviors. Our analysis therefore focuses on documenting emotional and physiological responses as potential precursors to safety-relevant decision-making.
4.1 GSR data
GSR was collected using a Shimmer3 GSR + sensor (sampling rate 128 Hz) with Ag/AgCl electrodes placed on the palmar surfaces of the middle and ring fingers. The GSR data collected during the VR activity provided a physiological marker of the participant’s arousal levels in the form of skin conductance data. Since skin conductance reflects the changes in the electrical properties of the skin, which are closely linked to emotional processes. GSR has been widely used as an index of physiological arousal in response to various stimuli (Boucsein, 2012; Cacioppo et al., 2007; Dawson et al., 2017). While this data does not provide conclusive evidence of specific emotional states beyond signaling arousal, GSR, in conjunction with other qualitative data collected for this study, can provide insight into how the participants were impacted. Therefore, the GSR data enabled an empirical test of whether the accident simulated in VR was successful in eliciting physiological arousal from the participants.
4.1.1 Data preparation
GSR was collected using a Shimmer3 GSR + sensor (sampling rate 128 Hz) with Ag/AgCl electrodes placed on the palmar surfaces of the middle and ring fingers. GSR data can be susceptible to artifacts caused by movement, environmental factors or signal drift over time. To preprocess the data, the raw GSR signals were first exported from the Shimmer3 device and imported into the Consensys software provided.
. Signals were zero-phase low-pass filtered using a 4th-order Butterworth filter (5 Hz cutoff) and then downsampled to 10 Hz. Motion artifacts were removed using standard amplitude/slope criteria (changes >0.05 µS within 100 ms or derivatives >10 µS/s), supplemented by visual inspection; flagged samples were linearly interpolated (Boucsein, 2012). Additionally, visual inspection of the data was conducted to identify and remove any sudden spikes or anomalies not associated with the experimental conditions.
A 20-second resting baseline prior to VR exposure was used for normalization. To synchronize the GSR data with the VR events, event markers inserted at the accident onset allowed the extraction of 10-second pre- and post-event windows for within-participant comparison. This was achieved by pressing a designated button on the Shimmer device at the exact moment the accident was triggered in the VR simulation. These event markers allowed us to segment the GSR data into “before-accident” and “after-accident” periods for analysis.
Next, to facilitate a within-participant comparison, the GSR data were trimmed so that equal durations of data represented the “before-accident” and “after-accident” periods. This approach ensured that any changes in skin conductance could be attributed to the accident simulation and not to other factors such as habituation or initial adjustment to the VR environment. The data segments were then baseline-corrected by subtracting the mean skin conductance level obtained during a rest period prior to the VR experience for each participant. This normalization accounted for individual differences in baseline arousal levels. Because the accident was a continuous multimodal stimulus, we analyzed baseline-corrected mean skin conductance level (SCL) rather than discrete SCR peaks.
The processed GSR data were then analyzed statistically to determine whether there was a significant difference in arousal levels before and after the simulated accident. This method provided an objective physiological measure of the emotional impact of the VR experience, complementing the self-reported emotional data from the surveys.
4.1.2 Hypothesis testing
As detailed in the Data Analysis section, a within-participant comparison of the GSR data before and after the accident event was conducted using the Wilcoxon signed-rank test, which is appropriate for analyzing non-normally distributed data. The Shapiro-Wilk test confirmed that the GSR difference scores were not normally distributed (W = 0.90, p < 0.001), justifying the use of this non-parametric test. The Wilcoxon signed-rank test assessed whether the median difference between the pre- and post-accident arousal levels was statistically significant, reporting the test statistic (V), standardized Z-score and exact p-value. Additionally, the rank-biserial correlation (r_rb) was calculated to quantify the effect size of the observed differences, following the reporting format used for the survey data in Table 1.
4.2 Post-VR interview
While the GSR data provided a physiological measure of the participants’ arousal levels, the post-VR interview was used to gain further qualitative insight into their reaction to the accident in the VR activity. These interview recordings were first transcribed using OpenAI's whisper API (Radford et al., 2022).
Following transcription, a thematic analysis was performed to systematically analyze the interview data, following the six-step approach outlined by Braun and Clarke (2006). The process began with familiarization, where the researchers thoroughly read and re-read the transcribed interviews to gain a deep understanding of the data. Next, initial codes were generated to identify meaningful segments of text that captured participants’ experiences and reactions to the VR simulation. These codes were then sorted and grouped into broader themes that reflected common patterns in the data (Nowell et al., 2017).
4.3 Survey responses
Because pre–VR ratings exhibited strong floor effects (a granularity trap), medians frequently collapsed to zero and masked meaningful within-participant changes. For descriptive clarity, we therefore report means and standard deviations, while using the paired Wilcoxon signed-rank test for inference, which is appropriate for ordinal data. For each emotion, we computed the mean difference with 95% bootstrap confidence intervals, the rank-biserial correlation (rrb) as the effect size and Holm-adjusted p-values to account for multiple comparisons.
5. Results
5.1 Descriptive statistics
In total, 55 participants took part in this study. Of these, 49 identified as male, while 6 identified as female, indicating a predominantly male representation in the sample. The participants in this study had an average work experience of 2.1 years in the construction industry. When asked about previous work-related injuries, 78% of participants responded that they had not been injured at work before. Conversely, 20% of participants reported having experienced work-related injuries in the past and 2% of participants declined to respond. For the participants who had been injured at work, the majority of injuries were classified as minor, requiring only first aid treatment. Only one participant reported experiencing a moderate-level injury, which necessitated medical attention but was not life-threatening in nature.
These descriptive statistics provide an overview of the participants' work experience, gender distribution and history of work-related injuries. The findings indicate that the participants had little work experience and most participants had not experienced an injury on a construction site before.
5.2 Significant impact on physiological arousal
The authors analyzed the GSR data to test whether the VR experience impacted participants' arousal levels. Because the accident was a continuous, multimodal stimulus rather than a discrete event, we analyzed baseline-corrected mean skin conductance level (SCL) rather than discrete skin conductance response (SCR) peaks, which are better suited for brief, punctuated stimuli. A Wilcoxon signed-rank test comparing mean skin conductance levels in the 10-second windows before and after the accident event revealed a statistically significant increase in arousal (V = 1,485, Z = 6.39, p < 0.001), with a rank-biserial correlation of r_rb = 0.87. This large effect size underscores the effectiveness of the VR simulation in eliciting physiological arousal.
The high effect size is especially relevant to this study as it indicates the VR construction accident simulation's effectiveness in eliciting physiological arousal, an internal state theorized to precede risk-averse decision-making. In other words, the VR experience had not just a statistically significant effect on arousal, but also a practically significant one, demonstrating its potential for engaging the emotional and physiological systems associated with safety-relevant appraisals. Whether this arousal translates to improved safety behavior in the field remains a target for future research.
5.3 Self-reported emotions questionnaire supports GSR findings
The self-reported questionnaire data gathered pre- and post-VR experience revealed substantial shifts in participants’ emotional states. While the physiological response (as indicated by GSR) does imply an emotional reaction, it does not tell us about the nature of that emotion. The questionnaire data provided more insight into the specific emotions that the VR experience evoked in participants. Table 1 reports the findings and the following sections highlight the major insights from the comparison of the self-reported emotions from before and after the VR experience.
Change in self-reported emotions after the VR experience
| Emotion | Mean pre (SD) | Mean post (SD) | Mean diff [95% CI] | padj (Holm) | Effect size rrb |
|---|---|---|---|---|---|
| Interested | 7.45 (1.66) | 7.35 (1.86) | −0.11 [−0.55, 0.33] | 1 | −0.085 |
| Distressed | 1.51 (0.96) | 2.45 (2.15) | 0.95 [0.39, 1.50] | 0.016 | 0.729 |
| Upset | 1.13 (0.39) | 2.04 (1.79) | 0.91 [0.48, 1.34] | <0.001 | 1 |
| Guilty | 1.24 (0.92) | 2.31 (1.90) | 1.07 [0.61, 1.53] | <0.001 | 0.931 |
| Scared | 1.42 (1.26) | 2.20 (1.98) | 0.78 [0.24, 1.32] | 0.026 | 0.7 |
| Hostile | 1.29 (1.08) | 1.47 (1.14) | 0.18 [-0.03, 0.39] | 0.529 | 0.538 |
| Enthusiastic | 6.62 (1.93) | 6.36 (2.44) | −0.25 [−0.85, 0.34] | 1 | −0.094 |
| Irritable | 1.49 (1.07) | 1.80 (1.74) | 0.31 [−0.15, 0.77] | 0.826 | 0.379 |
| Alert | 5.76 (2.28) | 6.58 (2.39) | 0.82 [0.28, 1.35] | 0.026 | 0.535 |
| Ashamed | 1.07 (0.33) | 2.35 (1.92) | 1.27 [0.78, 1.77] | <0.001 | 0.978 |
| Inspired | 5.62 (2.09) | 6.42 (2.27) | 0.80 [0.25, 1.35] | 0.056 | 0.485 |
| Nervous | 2.62 (1.89) | 2.51 (2.16) | −0.11 [−0.74, 0.52] | 1 | −0.072 |
| Determined | 6.24 (1.97) | 6.04 (2.07) | −0.20 [−0.69, 0.29] | 1 | −0.101 |
| Attentive | 6.56 (1.76) | 7.22 (1.89) | 0.67 [0.27, 1.06] | 0.024 | 0.542 |
| Active | 6.47 (2.08) | 6.91 (2.08) | 0.44 [−0.07, 0.95] | 0.298 | 0.352 |
| Afraid | 1.25 (0.82) | 2.09 (1.87) | 0.84 [0.37, 1.31] | 0.004 | 0.848 |
| Excited | 7.00 (1.86) | 6.27 (2.35) | −0.73 [−1.31, −0.14] | 0.178 | −0.41 |
Because pre-VR ratings exhibited strong floor effects (many responses clustered at 0–1), we report means and standard deviations for descriptive clarity while using Wilcoxon signed-rank tests for inference. Participants showed clear increases in several negative emotions after the VR experience, aligning with the arousal pattern observed in the GSR data. Distress rose from 1.51 to 2.45 (Δ = 0.95, 95% CI [0.39, 1.50], r_rb = 0.729), and feelings of being upset rose from 1.13 to 2.04 (Δ = 0.91, 95% CI [0.48, 1.34], r_rb = 1.000). Guilt and shame showed the largest changes: guilt increased from 1.24 to 2.31 (Δ = 1.07, 95% CI [0.61, 1.53], r_rb = 0.931) and shame increased from 1.07 to 2.35.
(Δ = 1.27, 95% CI [0.78, 1.77], r_rb = 0.978). Fear-related emotions also rose: “scared” increased from 1.42 to 2.20 (Δ = 0.78, 95% CI [0.24, 1.32], r_rb = 0.700), and “afraid”
increased from 1.25 to 2.09 (Δ = 0.84, 95% CI [0.37, 1.31], r_rb = 0.848). These findings are supported by paired Wilcoxon analyses with Holm-adjusted p-values and rank-biserial effect sizes, reinforcing the robustness of the observed emotional changes.
Several positive emotions showed modest declines. Enthusiasm decreased from 6.62 to 6.36 (Δ = −0.25, 95% CI [–0.85, 0.34], r_rb = −0.094) and determination decreased from 6.24 to 6.04 (Δ = −0.20, 95% CI [–0.69, 0.29], r_rb = −0.101), though neither change was statistically significant after Holm correction. These small shifts are consistent with the demanding and high-consequence nature of the VR scenario.
5.4 Themes from post-VR interviews
In addition to the physiological data, the semi-structured interviews revealed the qualitative aspects of how the participants reacted to the simulated accident and the VR experience in general. Specifically, 85% of participants (n = 47) reported experiencing strong emotions such as shock, sympathy or guilt after witnessing the accident. Many participants expressed feelings of responsibility for not preventing the accident, despite knowing that the scenario was simulated. This indicates that while the VR experience may not have fully created an illusion of being on a construction site, it successfully evoked realistic reactions like empathy and guilt, which could influence decision-making in real-life situations.
Other themes that emerged from the interviews included the realism and immersion of the VR environment, with 76% of participants (n = 42) commenting on the convincing nature of the simulation. Participants also noted the impact of the haptic feedback, with 65% (n = 36) stating that the physical sensations enhanced their emotional engagement and sense of presence. Because the design does not isolate haptic effects from general VR immersion, arousal should be interpreted as reflecting the combined multisensory experience.
Beyond the emotional reactions, other themes emerged out of the interviews, which are presented in Table 2 along with their description and example quotes from the transcripts.
Emergent themes from post-VR interview transcripts
| Themes | Descriptions | Quotes/Examples |
|---|---|---|
| Realism and Immersion | Most participants found the VR experience to be realistic and immersive, often forgetting they were in a simulated environment. The use of haptic feedback, realistic visuals, and environmental details contributed to the sense of immersion | “The virtual environment felt real, with realistic interactions among workers.” |
| Emotional Responses and Empathy | Many participants experienced strong emotional responses, such as shock and sympathy, when witnessing accidents in the VR environment. These emotions often led to an increased sense of empathy for virtual workers and a heightened awareness of the real-life implications of construction site accidents. Some participants felt a sense of responsibility for not preventing the accident, while others experienced distress or anxiety | “Felt distress and a sense of responsibility during the accident.” “I could sympathize with the injured worker.” The participant mentioned being able to sympathize with the virtual worker who got hurt, relating it to a real-life family member's experience |
| Learning and Awareness | Participants frequently reported that the VR experience increased their awareness of potential hazards on construction sites and improved their ability to identify and prevent accidents. The immersive nature of the VR experience was seen as an effective way to bridge the knowledge and experience gap, making participants more attentive to safety issues on real construction sites | Participant believed VR safety training would be impactful in increasing awareness on actual construction sites. “VR can contribute positively to safety training, making people more alert and aware of potential hazards.” |
| Comparisons to Traditional Safety Training | Some participants compared the VR experience to traditional safety training methods, such as classroom instruction or field experience. Many participants believed that the VR experience offered a more practical and interactive approach to safety training, allowing individuals to learn from mistakes without causing real damage | “The VR simulation differs from traditional safety classes, as it offers a more practical and interactive way of identifying hazards, rather than just learning about them through textbook material.” Participants felt the animations were realistic enough without being too graphic or cartoony |
| Potential Improvements and Limitations | Participants suggested various improvements to the VR experience, such as better human movement animations, more realistic visuals and incorporating personal backstories for virtual workers. Some participants also acknowledged limitations of the VR experience, such as the potential for a lack of empathy due to the knowledge that it was a simulation, and the reliance on the quality of the VR technology itself | “Some aspects of the VR world, such as lag and visuals, made it feel less realistic, but conversations among computer-generated characters added to the realism.” |
5.5 In-VR participant behavior
Although not an intended variable for measurement, researchers observed a noticeable lowering of participants’ vocal tone following the simulated accident. This was a qualitative observation noted by the research assistants rather than a systematically measured outcome. The tone change aligns with previous research on vocal indicators of emotional arousal and empathy. Such lowering of vocal tone has been associated with emotional changes (Banse and Scherer, 1996; Laukka et al., 2013; Ahn et al., 2013; Waal, 2008) and seriousness (Reicher et al., 2010) in psychological and linguistic literature. This may also indicate proactive behavior in response to the accident, which is a desirable outcome for a training experience (Michael and Chen, 2005). While no formal sound analysis was conducted, this anecdotal evidence suggests that participants were emotionally affected by the simulation, reinforcing the impact of the VR environment on emotional engagement.
As a noteworthy exploratory behavioral indicator, 22% of participants (n = 12) immediately reacted to the accident by holding out their hand or calling out to the virtual victim as though to warn them to get out of harm's way. Such instinctual behavior indicates the effectiveness of the simulation in facilitating a sense of presence on the virtual construction site (Slater, 2003; Slater et al., 1994). In addition, this may also indicate proactive behavior in response to the accident, which is a desirable outcome for a training experience (Michael and Chen, 2005).
6. Discussion
The present findings provide integrated physiological, emotional and behavioral evidence for the proposed mechanisms through which VR may influence safety-relevant decision processes. The observed increases in arousal (H1) and anticipatory emotions (H2) align with Mechanisms 2–3, while spontaneous protective behaviors and interview themes reflecting presence, empathy and responsibility (H3) correspond to Mechanisms 1 and 4. Taken together, this convergence supports a coherent pathway from immersive experience to internal state change that is theoretically linked to safety-relevant decision processes.
6.1 Comparison with prior construction training literature
The results directly engage with longstanding critiques of construction safety training. Traditional media such as static images, slides and 2D videos are repeatedly shown to lack ecological validity and emotional fidelity (Uddin et al., 2020; Eiris et al., 2021; Zuluaga et al., 2016). Our findings echo this literature, where participants demonstrated strong increases in several negative emotions following the VR experience, including distress, guilt, shame and fear-related emotions. These changes were statistically significant and showed large rank-biserial effect sizes (r_rb ranging from 0.70 to 0.98), reflecting levels of emotional engagement that are difficult to achieve with static images or 2D video. This aligns with (Albert et al., 2014b) and (Jeelani et al., 2020), who argue that immersive media better replicate perceptual cues and situational complexity necessary for effective hazard recognition. The observed GSR effect size (rank-biserial correlation = 0.87) further strengthens empirical work showing that immersive environments yield stronger physiological engagement than planar media (Eiris et al., 2020a, b).
6.2 Emotion-driven decision-making: theoretical expansion
The increases in emotions are theoretically significant. Decades of decision science demonstrate that emotions such as fear, distress and guilt are theorized to influence the shaping of risk perception and risk-averse choices independently of cognitive reasoning (Lerner and Keltner, 2000; Loewenstein et al., 2001). According to the somatic marker hypothesis (Damasio, 1996, 2004), emotional responses to stimuli serve as rapid, unconscious signals that guide decision-making, particularly in situations involving uncertainty and potential harm. When individuals encounter hazardous scenarios, negative emotional states such as fear and distress function as “somatic markers” that flag danger and bias decision-making toward risk-averse choices (Bechara and Damasio, 2005). Our findings are consistent with these theoretical accounts in a construction context: novices lacking vivid accident memories showed substantial emotional reactivity, consistent with theories of mental imagery and availability (Tversky and Kahneman, 1974; Miller et al., 1987; Corgnet et al., 2020). Participants' interview narratives referenced mental simulation of consequences such as “I felt responsible,” “It reminded me of a real accident” which map onto appraisal processes theorized to reflect internal states that precede safety-relevant decisions. We emphasize that these are proximal indicators; whether they predict actual reporting or intervention behavior requires longitudinal field validation (Zohar, 2000; Wang and Liao, 2021; Wang et al., 2021b; Joshi et al., 2021).
Physiological arousal findings from the GSR data, which showed a strong effect size (0.87), also conform to memory and attention mechanisms described by Cahill and McGaugh (1998), who identify emotional arousal as a driver of enhanced encoding for safety-critical cues. This is comparable to the effect sizes reported in other VR emotion elicitation studies. For example, (Rottenberg et al., 2007) report that film-based emotion induction often produces moderate physiological responses, suggesting that immersive VR can achieve comparable or stronger arousal due to greater sensory fidelity.
Our results suggest that VR may supply the sensory richness and vividness sufficient to evoke precursors associated with these pathways in novices. The addition of haptic feedback likely contributed to multisensory salience (Mechanism 3), consistent with prior haptic studies showing amplified immersion and emotional engagement (Basdogan et al., 2001; Boukhris and Menelas, 2017).
6.3 Protective behaviors and presence
Observation of spontaneous protective actions (22% of participants warned or reached toward the virtual victim) aligns with presence literature (Slater, 2003; Slater et al., 1994) and research indicating that embodied reactions reflect perceived realism. These immediate behaviors function as proximal indicators of intention to intervene (Mechanism 4). However, the relationship between emotional arousal and behavioral change is complex and not always linear. According to Witte (1992), fear appeals are most effective when they combine high threat with high efficacy. If individuals experience fear without a clear sense of how to respond effectively, they may engage in defensive avoidance rather than adaptive behavior change. Future iterations of the VR training should incorporate explicit skill-building components.
Some prior studies have attempted to measure behavioral transfer from VR training to real-world performance. Sacks et al. (2013) found that construction workers who completed VR safety training demonstrated better hazard recognition in subsequent site visits compared to control groups, though the effect diminished over time. Recent reviews (Moore and Gheisari, 2019) call for empirical evidence linking VR-induced emotional engagement to safety outcomes. This study offers such evidence, demonstrating significant alignment between physiological markers, subjective emotions and qualitative appraisals.
6.4 Extending VR literature
Finally, the study's results empirically validate the research-based approach employed by the researchers in their prior work to design construction-specific VR environments. By constructing the VR experience using prior findings on construction-specific VR design in a hazardous context (Patil et al., 2023a, b), the observed impact on emotional arousal confirms the theoretical frameworks and research findings that guided the design process. Additionally, Zuluaga et al. (2016) found that immersive training methods improved hazard recognition by approximately 15–20% compared to traditional classroom instruction, though they did not report standardized effect sizes or measure emotional responses. More recently, Yoo et al. (2023) reported that VR training significantly increased risk perception (β = 0.34, p < 0.01) and training satisfaction (β = 0.42, p < 0.01), mediated by telepresence. Our findings complement this work by providing direct physiological evidence of emotional arousal, rather than relying solely on self-reported risk perception. Establishing the link between these precursors and downstream safety behavior remains an important direction for future research.
7. Limitations and future work
While this study provides valuable insights into the potential of VR experiences to elicit emotional arousal among construction novices, several limitations must be noted. Our use of CM students limits generalizability to practitioner populations with greater field tenure. Although construction-professional input informed scenario design, a direct practitioner replication is needed. A planned follow-up with field crews and supervisors will help estimate effects in experienced populations and test whether prior accident exposure moderates arousal, appraisal-related language and proximal action.
The study measured the immediate impact of the VR experience on emotional arousal using physiological and self-reported data, but it did not directly assess whether these heightened states translate into improved safety behaviors on actual construction sites. The connection between emotional arousal and subsequent safety behavior therefore remains hypothesized. Future research should evaluate whether the emotional responses elicited by VR training have lasting effects on hazard recognition and risk-averse behaviors, ideally through longitudinal follow-up or real-world field studies.
Because the study did not include a haptics-on/off comparison, a VR-only control condition or counterbalancing, we cannot isolate the specific contribution of haptic feedback from general VR immersion or novelty effects. Elevated arousal may reflect the combined influence of visual immersion, embodied presence, novelty and haptics rather than haptic cues alone. Accordingly, we have tempered causal claims about the unique effect of haptic feedback and interpret physiological arousal as arising from the multisensory VR experience as a whole.
Sample diversity presents another limitation. Participants were primarily third- and fourth-year CM students from a single university, recruited through convenience sampling. This relatively homogeneous pool may not represent the wider range of novices or workers entering construction with varied educational, cultural and experiential backgrounds. Broader recruitment would enhance external validity and provide a more comprehensive understanding of VR training impacts.
Finally, we did not collect data on prior VR experience, motion-sickness susceptibility or safety-related coursework. These factors may moderate emotional arousal, presence and hazard recognition performance. For example, VR-novice participants may experience heightened arousal due to novelty, while individuals with high cybersickness susceptibility may show elevated GSR unrelated to hazard content. Likewise, students with more extensive safety coursework may have different appraisal responses. Future work will explicitly measure these characteristics and consider them as covariates or grouping variables.
In conclusion, while this study demonstrates the potential of VR simulations to elicit emotional arousal in construction novices, further research is necessary to establish a direct link between such emotional engagement and actual improvements in safety behaviors. By addressing the limitations identified, future studies can build upon these findings to develop more effective and evidence-based VR training programs that not only engage users emotionally but also lead to tangible improvements in safety outcomes on construction sites.
8. Conclusion
This study shows that VR with haptic feedback can elicit anticipatory emotional states and physiological arousal, internal precursors that are theoretically linked to safer decision processes, in novices. By increasing fear, distress and guilt in response to an accident scenario and by evoking protective behaviors in the moment, the simulation targets precursors of reporting and intervention. We conclude that VR influenced internal precursors (arousal, anticipatory emotions, protective reactions) that are theorized to precede risk-averse action.
The physiological evidence from GSR data, paired with self-reported emotions and interview insights, confirms that the VR experience not only engages users at affective and physiological levels theorized to influence risk perception and safety-related appraisals. Participants exhibited strong emotional reactions such as shock, empathy, guilt and a sense of responsibility – emotions that are essential for understanding the consequences of safety lapses. These findings suggest that VR can bridge the experiential gap for novices lacking real-world experience with construction hazards by eliciting emotional and physiological responses that may support the development of risk-averse dispositions. Whether these precursors translate to improved safety behavior in the field is a critical question for future longitudinal research.
Future research should explore the long-term impact of such emotional engagement on actual safety behavior in the field. Longitudinal studies would provide crucial insights into whether VR-based training translates into sustained improvements in hazard recognition and risk mitigation. Additionally, further investigation into the role of haptic feedback and other immersive technologies will help optimize the design of future VR training programs.
Ethics statement
This study was approved by the Arizona State University Institutional Review Board (IRB ID: STUDY00007496, approved January 5, 2018) under expedited review categories (6) Voice, video, digital or image recordings and (7) (a–b) Behavioral and social science methods. All participants provided written informed consent prior to participation using the IRB-approved consent form, which described study procedures, potential risks, confidentiality protections and the voluntary nature of participation.
Participant compensation: Students who took part in the study received course extra credit as an incentive for participation. To receive the extra credit, students were required to complete the VR activity and submit a written report on hazard recognition. Participation was entirely voluntary, with alternative opportunities for equivalent extra credit available to those who chose not to participate.
Risk mitigation during VR exposure: Participants were briefed on potential VR-related discomfort (e.g. motion sickness, disorientation) and informed they could remove the headset and discontinue at any time without penalty. A researcher was present throughout all VR sessions to monitor participant comfort and safety.
Debrief and safety procedures: Following the VR experience, participants were debriefed about the study objectives and encouraged to discuss their reactions. Those experiencing discomfort were provided time to recover before leaving and were given contact information for the research team and the ASU IRB.
Data protection: All physiological and questionnaire data were de-identified immediately upon collection and stored on password-protected, encrypted servers accessible only to the research team. All procedures adhered to the ASU INVESTIGATOR MANUAL (HRP-103) and the university's data-security and confidentiality requirements.
