1. Introduction
The construction industry often requires several workers to perform hands-on tasks to produce a final product. This makes construction workers especially vulnerable to occupational accidents, such as falling or electrocution [
1]. According to a report from the Korean Ministry of Employment and Labor (MoEL), the construction industry recorded 9.41 accidents per 1000 workers in 2018, while the average rate for all industries was only 5.36 [
2]. Furthermore, according to Zhao and Lucas [
3], 80% of the occupational accidents in the construction industry are caused by human error, with at least 49% being due to unawareness or misjudgment, meaning that they are preventable.
Over the years, various efforts to improve the safety of construction workers have been made, one of them being the implementation of mandatory safety training. Since 2012, construction workers in South Korea have been obligated to complete a four-hour basic safety and health training. The training focuses mainly on introducing various hazards in a construction site and ways to prevent accidents from happening therein. The workers must also complete additional training every time they change workplaces or when they operate special equipment, such as excavators.
However, despite these training sessions and their content, the current safety training has not resulted in an improvement in accident rates in the construction industry. According to the Korean MoEL, the construction industry exhibited an increase in the number of accidents per 1000 workers from 7.48 in 2015 to 9.41 in 2018, in contrast with the downward trend of accident rates in other industries [
2]. The increase in accident rate implies that the knowledge and skills garnered from the safety training are not being transferred to the workplace.
To enhance the effectiveness of safety training and promote safe working practices of construction workers, different training methods and technologies are being employed and tested with the current safety training. For example, advances in the IT sector have promoted the use of e-learning as a tool for safety training. Compared with the traditional training method, the use of e-learning has reduced training costs dramatically while the quality of the training has improved [
4].
Recently, the e-learning trend has further advanced with the adoption of virtual reality (VR) technology. VR safety training allows construction workers to interact in a virtual environment using a head-mounted display and a controller [
3,
5]. This allows them to experience various workplace hazards and practice safe working procedures in a safe, controlled environment [
6,
7]. In addition, a systematic review of literature on VR/AR use in the educational context found that the use of VR/AR technology enhances interactivity, encourages collaborative learning, and increases learner satisfaction [
8]. Furthermore, owing to the COVID-19 pandemic, the use of VR as a training tool is gaining attention because it can be conducted with minimal human contact [
9].
Although the adoption of VR safety training in the construction industry is rapidly increasing, studies on understanding the factors that increase VR safety training effectiveness are still lacking. Some studies have focused on developing a VR training system and comparing its effectiveness to that of other training methods using ANOVA or a paired sample
t-test as analysis methods; nevertheless, simply demonstrating that the VR method is more effective fails to explain the reason for its effectiveness. For example, Sacks et al. [
10] found that trainees who completed VR safety training showed higher test scores for certain construction tasks compared with those who received traditional classroom training; however, the researchers failed to explain the reason for the effectiveness of the VR-trained group in selective tasks. This type of methodological approach to understanding effectiveness has led researchers to assert that research on training effectiveness lacks a theoretical basis [
11].
Therefore, this study aims to employ a more theoretical approach to explain VR construction safety training effectiveness. Specifically, we utilize the technology acceptance model (TAM) to determine the specific features of VR safety training and risk perception toward occupational accidents that affect the workers’ perception toward VR safety training and finally, training transfer. Consequently, we aim to provide timely implications for improving the effectiveness of construction safety training using VR.
3. Research Model and Hypotheses
Prior studies on VR safety training have focused mainly on comparing the effectiveness of the VR training method with that of traditional methods. However, simply comparing the performance between different training groups fails to explain the reason for the effectiveness of one training method over another. This has been a major issue in studies related to measuring training effectiveness. Clark et al. [
31] specifically mentioned that “without a theoretical basis for studying the techniques … researchers are often at a loss either to explain why they are effective or to predict their effectiveness in other settings”.
Therefore, this study aims to provide an explanation for the effectiveness of the use of VR as a safety training tool. Specifically, by applying the TAM, we explain the effect of the characteristics of VR construction safety training on trainees’ beliefs regarding VR safety training, which affect the trainee’s satisfaction and the training effectiveness. Our research model is presented in
Figure 1.
3.1. Modified TAM
The TAM posits that an individual’s belief, specifically their perceived usefulness and perceived ease of use concerning a certain technology, affects their attitude, usage intention, and acceptance of the technology. The two cognitive beliefs presented in TAM are perceived usefulness and perceived ease of use, which are each defined, respectively, as “the degree to which a user believes that the use of a certain system would enhance their performance” and “the degree to which a user believes that using a certain system would be free of effort [
16]”.
Prior studies have confirmed the effectiveness of TAM when explaining and predicting acceptance of a newly emerging technology [
32,
33]. In a training or education context, TAM has been utilized to study an individual’s intention to use e-learning or other newly introduced training tools. For example, Chang et al. [
34] adopted the TAM to study the factors affecting students’ intention to use e-learning. They found that although perceived ease of use had no significant effect on perceived usefulness, both perceived usefulness and perceived ease of use positively affected students’ intention to use e-learning. Additionally, Wu and Vu [
35] confirmed the positive effect of perceived usefulness and perceived ease of use on an aviation student’s intention to use an augmented reality maintenance training system.
Additionally, in cases where technology adoption is mandatory, satisfaction was found to be an effective indicator of user acceptance. For example, to study the acceptance of web-based training among construction professionals, Park et al. [
24] utilized user satisfaction as a variable replacing usage intention in the TAM model. Similarly, Nah et al. [
36] applied ‘symbolic adoption’ as an alternative to usage intention regarding the acceptance of an enterprise system, and measured the construct through items such as ‘I am excited about using the SAP system in my workplace’.
Therefore, based on prior studies, we defined perceived usefulness as the degree to which a trainee believed that the VR safety training would enhance their performance, and perceived ease of use as the degree to which a trainee believed that the VR safety training would be free of effort. Furthermore, based on studies employing the TAM as a theoretical basis for studying the acceptance of technologies in a mandatory setting, we present the following hypotheses:
H1. Perceived usefulness has a positive effect on training satisfaction.
H2. Perceived ease of use has a positive effect on training satisfaction.
H3. Perceived ease of use has a positive effect on perceived usefulness.
3.2. External Variables of TAM
3.2.1. Telepresence in VR
Telepresence is a multi-dimensional concept consisting of vividness and interactivity [
25]. While vividness is determined by the sensory breadth and depth perceived by the user, interactivity is determined by the response speed, range, and mapping of the VR technology. In summary, higher levels of telepresence could indicate that the users perceived the virtual environment as reflecting the real world. Prior studies have utilized these two dimensions of telepresence to analyze its effect on flow experience or satisfaction [
28].
Furthermore, recent studies focusing on the acceptance of VR or AR have utilized TAM, and they have observed the effect of telepresence on user belief. For example, a study on the acceptance of AR interactive technology, in which consumers were allowed to try on clothing products using AR, confirmed the significant effect of presence on the perceived usefulness and ease of use for both consumers with high and low cognitive innovation [
37]. Similarly, Oh and Yoon [
38] confirmed the effect of presence on perceived usefulness and perceived ease of use when adopting haptic enabling technology (HET), which allowed users to control a virtual environment using touch.
Therefore, to study the effect of telepresence on user beliefs in the VR construction safety training context, we present the following hypotheses:
H4(a). Vividness has a positive effect on perceived usefulness.
H4(b). Vividness has a positive effect on perceived ease of use.
H5(a). Interactivity has a positive effect on perceived usefulness.
H5(b). Interactivity has a positive effect on perceived ease of use.
3.2.2. Risk Perception: Perceived Vulnerability and Severity
Perceived vulnerability and perceived severity are each defined as an individual’s perceived probability of a threatening event happening to them and the severity of the damage the event will cause to their life or health, respectively. In this study, these two constructs are defined as the probability of an accident occurring to the construction worker at their workplace, and the damage the accident would do to their health.
Perceived vulnerability and perceived severity are often used in research to study an individual’s protective behaviour toward a certain threat. They are also often integrated with TAM as external variables affecting an individual’s perceived usefulness toward a health-related technology. For example, Ahadzadeh et al. [
39] found that perceived health risk, which is a multi-dimensional construct composed of perceived severity and perceived susceptibility, has a positive effect on perceived usefulness in the context of health-related internet use. Furthermore, Hansen et al. [
40] found that perceived risk significantly affected perceived usefulness when studying consumers’ intention to use social media for transactions. Dou et al. [
41] also confirmed the effect of perceived health threat on perceived usefulness in a study identifying the factors affecting the acceptance of M-health technologies.
Because construction workers are especially vulnerable to life-threatening accidents in the workplace, the level of risk perceived by the worker could have a significant effect on their perceived usefulness of VR safety training. If they consider themselves to be at risk of accidents, there is a higher chance they will think that the training content would be useful in keeping them safe. Therefore, the following hypotheses are presented:
H6. Perceived vulnerability has a positive effect on perceived usefulness.
H7. Perceived severity has a positive effect on perceived usefulness.
3.3. Training Effectiveness
Training transfer is one of the most widely used criterion for measuring training effectiveness. In several studies, training satisfaction was confirmed to be a significant predictor of training transfer [
42,
43]. Therefore, by integrating the TAM with the concept of training transfer, the following hypothesis is presented:
H8. Training satisfaction has a positive effect on training transfer.
Figure 1.
Proposed Research Model.
Figure 1.
Proposed Research Model.
5. Results
We conducted partial least squares structural equation modelling (PLS-SEM) using the SmartPLS 3.0 program. PLS-SEM has advantages over traditional analysis methods because it allows for the assessment of measurement error and can predict latent variables using observed variables. In addition, it effectively analyses the causal relationship between several complex variables. Furthermore, compared to CB-SEM, PLS-SEM shows greater statistical power and is applicable when the structural model is complex [
54]. Finally, the PLS-SEM method can analyse ordinal data, such as survey data, based on a Likert scale; this was also performed in our study [
55].
5.1. Measurement Model Testing
Prior to hypotheses testing, we verified the measurement items by confirming their convergent and discriminant validity. Convergent validity is confirmed when (1) the factor loadings of the survey items exceed 0.7, (2) the average variance extracted (AVE) of the constructs exceeds 0.5, and (3) the composite reliability (CR) and Cronbach’s alpha exceed 0.7 [
54]. Furthermore, discriminant validity is confirmed if the heterotrait–monotrait ratio of correlations (HTMT) for each construct is lower than 0.9 [
54]. As shown in
Table 3 and
Table 4, all the criteria for convergent and discriminant validity were met, confirming the validity of our measurement model.
Furthermore, we tested for possible collinearity issues regarding our structural model. As shown in
Table 5, all VIF values were less than 3, confirming that collinearity is not an issue [
54].
5.2. Hypotheses Testing
We tested our hypotheses using the PLS-SEM method. A bootstrap resampling procedure was performed to test the significance of the paths. The results, including the path coefficients and the overall explanatory power, are shown in
Table 6.
The results of the hypotheses test showed that the effects of interactivity and perceived vulnerability on perceived usefulness were statistically insignificant at p < 0.05. Thus, H5(a) and H6 were rejected. However, all the other hypotheses were statistically significant and thus supported. Vividness (β = 0.160), perceived severity (β = 0.155), and perceived ease of use (β = 0.621) had a significant effect on perceived usefulness, whereas vividness (β = 0.303) and interactivity (β = 0.255) significantly affected the perceived ease of use. Moreover, both perceived usefulness (β = 0.403) and perceived ease of use (β = 0.254) significantly affected training satisfaction, which led to training transfer (β = 0.698).
Additionally, while H1, H2, H4(a), H4(b), H5(b), and H7 showed weak effect size (f^2 ≥ 0.02), H3 and H8 showed strong effect size (f^2 ≥ 0.35) [
56]. The specific f^2 value of each hypothesis along with its significance is presented in
Table 6.
Finally, the explanatory power (R^2) of perceived usefulness (R^2 = 0.559), perceived ease of use (R^2 = 0.258), training satisfaction (R^2 = 0.374), and training transfer (R^2 = 0.487) all exceeded the required threshold of 0.10, proposed by Falk and Miller [
57].
7. Conclusions
This study aimed to elucidate how the use of VR helps increase the effectiveness of safety training in the construction industry. We proposed a research model, based on the TAM, to examine the impact of technology- and usage-specific factors of VR safety training on the effectiveness of training. A PLS-SEM analysis of 248 construction workers in Korea was conducted. The results supported most of our hypotheses. Vividness, interactivity, and perceived severity had a significant effect on trainees’ beliefs and attitudes concerning the VR safety training. Furthermore, the trainees’ satisfaction regarding the training was crucial in increasing the transfer of the skills and knowledge obtained via safety training.
Our findings provide several meaningful implications for improving the effectiveness of a VR-based construction safety training. First, creating an immersive virtual environment that creates a feeling of telepresence is critical for enhancing training effectiveness. This could be achieved by providing various sensory stimulations through the use of haptic suits and other related technologies. Also, an interactive virtual environment with a high degree of freedom must be created. Finally, an effort to keep trainees interested and satisfied must be made by providing updates based on trainee feedback.
Although our study provided meaningful implications, it is not without its limitations. First, our survey was only conducted in Korea. Different countries have different training contents, work practices, and safety regulations. Thus, comparing different cultures may provide a more generalized result and new implications. Another shortcoming of our study is that we analysed a limited number of samples only consisting of construction workers. Future studies could conduct a multi-group analysis between workers and managers, or between construction tasks. Finally, our study fails to consider the possible bias caused by the novelty effect. To mitigate the novelty effect, future studies may need to provide tutorial courses so that the trainee is familiar with the VR before the training begins [
64]. Also, a time-series analysis on training effectiveness may provide new implications while eliminating possible bias caused by the novelty effect.