The Design of a Posture Instruction Atlas and the Prevention of Construction Workers’ Work-Related Musculoskeletal Disorders (WMSDs): A Study on Attention Allocation and Cognitive Load Based on Eye Tracking

Abstract

Training construction workers in safe postures for their tasks could help them avoid unsafe postures and reduce work-related musculoskeletal disorders (WMSDs). This study compared two forms of atlas design in facilitating workers’ learning postures, including their differences in guiding workers’ attention allocation and cognitive load during the learning process. One kind of atlas graphically shows the correct postures to perform construction tasks, and the other adds wrong demonstrations alongside the right ones. Eye-tracking technology was utilized to measure attention allocation and cognitive load. An experimental study was conducted, with 52 construction workers being invited as participants. The results indicated that workers significantly distributed more attention to diagrams than texts and more attention to diagrams showing execution postures than preparatory postures. Moreover, the workers had significantly longer fixation durations on the key body parts when there were wrong demonstrations, which ultimately improved their learning outcomes. There were no significant differences in cognitive load. Suggestions for designing an instructional atlas to enhance construction workers’ occupational health education can be obtained from the findings, including applying diagrams more instead of texts to describe how to correctly perform construction tasks, emphasizing the importance of preparation posture when performing construction tasks, and adding wrong demonstrations showing consequences, with visual cues being positioned on the key body parts.

1. Introduction

Musculoskeletal disorders (MSDs) are a group of painful disorders of soft tissues (muscles, tendons, nerves, joints, cartilage, and ligaments) [1]. For construction workers, the prevalence of work-related musculoskeletal disorders (WMSDs) is very high worldwide [2]. What is worse, WMSDs are the leading cause of disability among construction workers [3]. Symptoms of WMSDs include pain, aching, discomfort, numbness, tingling, and swelling that normally occurs in the back, shoulders, neck, legs, wrists, fingers, elbows, and arms. These disorder issues have led to a shift of labor in the construction industry to other sectors, which seriously affects the sustainable development of the construction industry [4].

There are many well-recognized WMSD risk factors in construction workplaces due to the highly physically demanding tasks, including repetitive motion, high-force exertion, awkward body posture, vibration, and contact force [5,6]. Among these factors, awkward body postures refer to positions of the body segments that deviate from a neutral posture, such as back bending or twisting to lift materials, overhead work, stooping, and squatting [7]. An awkward posture can result in higher muscle activation and muscle overloading because it leads to less efficient force production in the skeletal muscle compared to a neutral posture [8]. Performing activities in awkward postures and for an extended period will pose a lot of strain and fatigue, most commonly to the back, knee, and neck or shoulders [3,9].

The construction industry takes some measures to reduce workers’ awkward postures to avoid WMSDs. One typical method is using observation-based techniques like the rapid upper limb assessment (RULA), Ovaco working postures assessment system (OWAS), posture, activity, tools, and handling (PATH), and rapid entire body assessment (REBA) [4] to observe and rate workers’ awkward postures on-site or off-site (i.e., through video recording). However, the observation methods are time-consuming and are potentially biased due to the subjective judgment of the raters [10]. Moreover, advanced information technologies have also been applied for automated posture monitoring [11], e.g., Yan et al. [12] attached wireless inertial measurement unit (IMU) sensors to workers’ heads and backs to infer the angles described by the head, neck, and trunk, which can serve as a warning and thus prevent WMSDs. Nevertheless, applications of this method on job sites would be restricted by the battery capacity of the IMU sensors and their accuracy in identifying body postures. Both observation-based techniques and automated posture monitoring are relatively passive approaches, relying on observing or monitoring workers’ posture. Instead, teaching workers safe postures is a relatively proactive approach. It has also been demonstrated that workers’ awkward postures can be prevented by adequately training construction workers to make them aware of safe and unsafe postures for their tasks [3,13] to help them recognize and avoid unsafe postures.

The use of atlases, i.e., using pictures to teach construction workers the correct working posture, is an appropriate approach for training workers about safe postures. Atlases vividly presenting postures can lead to more effective information communication and presentation through supporting human cognition [14]. Although applications of visual aids (i.e., atlas) are helpful in the occupational safety and health training of construction workers, it must be noted that the assistance of visualization in human cognition is influenced by the design of the atlas itself (i.e., the representations of visualization). For example, the cognitive fit theory suggests that the fit between the information presentation format and the task can influence problem-solving performance [15]. If inappropriately designed visualizations (i.e., atlases) are adopted, it would overwhelm the viewer and undo the benefits of cognitive support [16,17]. However, there are few studies in the construction research community that have focused on the effects of visual aids’ (e.g., atlas) design on workers when the visual aids are adopted as a support tool in occupational safety and health training in the workplace. Therefore, we proposed conducting preliminary research on the effects of visual aids’ (i.e., atlases) design on supporting the training of educating construction workers’ safe working postures during construction tasks. The findings could indicate the relationship between the visualization design and the effectiveness of occupational safety and health training and provide insights for developing appropriate instructions for training. Specifically, this study intends to compare the differences between the two forms of atlas design in facilitating workers learning working postures. One graphically shows the correct postures to perform construction tasks, and the other also adds wrong demonstrations with consequences alongside the right ones because displaying the consequences in the case of accidents was demonstrated to enhance trainees’ understanding of the visualizations (e.g., [18]). How workers allocate their attention and how much mental effort they put (i.e., cognitive load) during the learning process were selected to indicate the cognitive supports of the instructional atlas.

Attention allocation and cognitive load were chosen for the investigations because they are fundamental during learning. According to Wickens’ model [19] (Figure 1), a human general information-processing model describing the cognitive processes when interacting with external information, there are four cognitive stages: sensory processing, perception, decision, and response selection and response execution. Firstly, subjects use senses (e.g., eyes, ears, etc.) to process the external stimulus (i.e., sensory processing). Next, subjects select a part of the sensory information for further perception, which involves determining the meaning of the sensory information. After perception, it would trigger an immediate response and/or retain the information and then go to the response selection. Attention is a kind of cognitive resource that is required by each cognitive activity, which plays two key roles during this process. Its first role is a filter of information which selects certain elements of the stimulus to be further perceived. In the second role, it acts as a fuel that provides mental resources to the various stages of information processing. Regarding how workers learning from the instruction alters (i.e., a visual–perceptual task), information is selected by visual attention, the conscious devotion of attention to visual information required to absorb it and transfer it to working memory. Therefore, focusing on relevant information and ignoring irrelevant information (i.e., attention allocation) is important for improving learning outcomes or task performance [20]. Moreover, since human’s attentional resources are limited, the instruction needs to be designed to make optimal use of these limited processing resources [21].

The working memory provides a mental workspace to store and process information. When what workers learn from the instruction alters, all information needs to be processed in the working memory before it can be integrated with prior knowledge and stored in the long-term memory [19]. However, the capacity of a human’s working memory is limited; if the cognitive demands of a particular task (e.g., learning from the instruction alters) put a big burden on working memory or are even beyond its capacity, the learning outcomes would be negatively impacted [21,22]. Hence, the cognitive load associated with learning has also been regarded as the major factor determining the success of instructional methods [22]. To obtain better learning results, instructional materials should be presented in alignment with cognitive processes to allow more memory resources for knowledge acquisition [16]. The way in which information is presented (e.g., stimulus design) can affect the amount of load placed on the working memory system and thus affect performance [23].

To conclude, this study will compare workers’ attention allocation and cognitive load while learning working postures from two kinds of instructional atlases with different designs, where eye-tracking will be applied to measure attention and cognitive load. The findings will indicate the relations between the visualization design and the effectiveness of occupational safety and health training among construction workers and provide insights for designing appropriate atlases for teaching workers safe postures when performing construction tasks.

2. Materials and Methods

The study was implemented following the framework in Figure 2. First, an experimental study was carried out whereby the subjects were required to learn safe postures from the instructional atlases, and there is a screen-based eye tracker recording their eye gaze and eye movement. During the experiment, the subjects’ attention allocation, cognitive load, and task performance, which contain multiple metrics, were collected. Finally, these metrics were applied for statistical analysis. Detailed descriptions of the research approach are in the following sections.

2.1. Experiment Design

2.1.1. Participants

Construction workers (n = 52) were invited to join this study, aged 20 to 25 (22.5 ± 1.5), with less than three years of construction experience. Their work types include 9 carpenters, 20 masons, 10 electricians, 6 roofers, 3 plumbers, and 4 pipefitters. The sample size was determined via prior power analysis in GPower 3.1.9.7 Software [24], with a power over 0.800. All participants had normal or corrected-to-normal vision. They were asked to get enough sleep before the test and were recommended not to smoke or use caffeine or other stimulants 24 h before the experiments. All participants were paid to enhance their motivation and to encourage them to perform the experiments seriously and correctly. The study was approved by the Ethical Research Committee at Hunan Normal University. Moreover, informed written consent was obtained from each participant.

2.1.2. Stimulus

Each set of stimuli represents the same construction task and consists of two stimuli (S1 and S2) that differ in design. One graphically shows how to complete the task using the correct postures (Figure 3A), and the other demonstrates the consequences of incorrect postures (Figure 3B). There are six sets of stimuli in total, covering a series of tasks: bending down to lift heavy objects (long boards of lumber, sheet goods, a rebar, a concrete block, packing cement, and a cement mortar drum) from the floor, which might lead to lower back injuries. The lower back is one of the most commonly affected body parts among construction workers and is a leading cause of disability [25]. The subjects selected for this study also performed these tasks frequently in their daily routines.

Specifically, as shown in Figure 3A, the stimulus shows merely correct postures (S1) consisting of four sub-diagrams (one starting position and three execution postures). There are also explanatory texts below the diagrams. The other stimulus (S2) adds an incorrect demonstration next to the correct sub-diagram (Figure 3B), displaying possible lower back injuries using visual cues.

The hierarchical task analysis (HTA) method [26] was applied for depicting the movements of the workers when performing a specific task. HTA is treated as a strategy for examining tasks, aimed at understanding task contexts [26]. Specifically, HTA was used to divide the overall working task into sub-tasks, elementary tasks, and specific activities following the hierarchical manner by means of a tree diagram [27], as shown in Figure 4A. The breakdown of the construction task in Figure 3 is shown in Figure 4B. The first level refers to the task of lifting lumber. The second level consists of the four sub-tasks. The first step is the preparation posture, and the next three steps are the execution postures, namely, lifting lumber from the ground, then lifting it onto one’s shoulders, and finally lifting and walking. The third level shows the specific posture for each step of the sub-task. The first preparatory posture includes spreading the legs and squatting; lifting lumber from the ground requires lifting the board up by using the power of the legs; lifting onto the shoulders requires the workers to lift the item onto the shoulders and to keep changing the shoulder side; and the last execution posture includes two postures, i.e., lifting and walking. The final level is the matters that need attention when performing a posture. During the first sub-task, the workers must keep their backs straight while squatting. Regarding the second sub-task, the workers are required to keep their back straight when lifting the lumber up. As for the third sub-task, it is necessary to change the side of the shoulders. While lifting lumber and walking, the workers must keep their waists upright.

2.1.3. Apparatus

Eye-tracking technology was utilized to measure attention allocation and cognitive load in this study. Eye-trackers measuring the point of eye gaze and eye movement have been widely used for human cognition studies. Visual attention is closely linked to where a person is looking, which can be captured by eye-tracking [28]. Eye-tracking has already been suggested as an effective tool for assessing attention allocation by researchers from various disciplines [29,30]. In addition, eye-tracking also provides metrics such as pupil diameter changes and fixation duration [31] to indicate subjects’ cognitive load. More importantly, eye-tracking is non-intrusive, which can reduce effects on subjects’ task performance. It can also indicate the subjects’ cognitive process in processing visualizations, helping objectively understand people’s feelings towards visualization and explaining specifically what causes performance problems or improvement [32]. In this study, the applied eye-tracker was Tobii Pro Fushion (Tobiitech, Stockholm, Sweden), with a sampling rate of 250 Hz. It was attached to the bottom edge of a laptop screen to track and record the participants’ eye movements.

2.1.4. Experimental Procedure

Each subject joined the experiments twice. They answered ten questions the first time to indicate their prior knowledge about working postures. They were then randomly divided into two equal groups (G1 and G2) the second time, with one group (G1) receiving the stimuli type S1 and the other (G2) receiving S2.

During the experiment, the participants sat approximately 30 cm to 60 cm from the laptop screen. The stimulus was presented on a laptop (15.6 in., resolution: 1920 × 1080), and the experimenter could monitor the participants’ real-time screen through an external screen connected to the laptop. Tobii Pro Lab software (V.1.152, Tobiitech, Stockholm, Sweden) was used to control the stimulus presentation, with Tobii’s five-point for calibration. It was accepted if all five points for the two eyes were caught by the Tobii, with smaller than 0.5 degrees of the visual angle.

Before the formal experiment, the participants received a briefing of the experiments and practiced an experimental demo to ensure they understood the experimental tasks. In the formal experiment, the participants were suggested to keep their body and head still to ensure the eye-movement data quality, and they followed the procedure, as shown in Figure 5. Firstly, there was an instruction informing the participants of the construction task to be presented. The participants were given 5 s to read it through. Then, after a 1000 ms blank view, the stimulus appeared without time limits, and the participants could switch it by pressing any keyboard button. After that, the instruction for the next stimulus appeared. After observing all six stimuli, there was a test for examining their learning outcomes, with 18 single choices (6 were about starting position and 12 were about execution postures) and a time limit of 15 min. A single choice contains four items. One option was designed as “unsure”, and the participants were required to choose it when they failed to find the answer. This option was to avoid the participants guessing the answer. Finally, the subjects filled out a questionnaire after finishing the test. They were asked where they allocated more attention between two elements and to choose reasons or report their reasons, as shown in

Table 1. Design of questionnaires after finishing the test.

Index	Questions	Available Items for Reasoning
1	Which one did you pay more attention to, diagrams or texts, and why?	It is easier for me to understand. It attracts my attention more. I prefer to observe it. Others.
2	Which kind of sub-diagrams did you pay more attention to, starting position or execution postures, and why?	I think it is more important. I think the other one is too easy. There is more I would like to know. Others.
3	Which one did you pay more attention to, diagrams showing correct postures or those showing incorrect postures, and why? (only for G2)	I need to put more effort into understanding it. I think it is more important. Others.
4	Could you notice the key body parts?	-
5	Were you attracted by the visual cues? (only for G2)	-

. They also filled out the NASA-TLX questionnaire, a standard questionnaire for subjects self-reporting their mental workload [33], to report their workload.

2.2. Data Collection and Analysis

Task accuracy was collected to indicate the participants’ task performance. The following metrics were used to measure the subjects’ attention allocation within the stimuli.

$$\mathrm{R}1=\frac{\mathrm{Fixation} \mathrm{duration} \mathrm{in} \mathrm{per} \mathrm{unit} \mathrm{area} \mathrm{of} \mathrm{texts}}{\mathrm{Fixation} \mathrm{duration} \mathrm{in} \mathrm{per} \mathrm{unit} \mathrm{area} \mathrm{of} \mathrm{diagrams}}\times 100\%$$

(1)

$$\mathrm{R}2=\frac{\mathrm{Fixation} \mathrm{duration} \mathrm{in} \mathrm{per} \mathrm{piece} \mathrm{of} \mathrm{starting} \mathrm{positions}}{\mathrm{Fixation} \mathrm{duration} \mathrm{in} \mathrm{per} \mathrm{piece} \mathrm{of} \mathrm{excution} \mathrm{postures}}\times 100\%$$

(2)

$$\mathrm{R}3=\frac{\mathrm{Fixation} \mathrm{duration} \mathrm{in} \mathrm{the} \mathrm{key} \mathrm{body} \mathrm{parts}}{\mathrm{Fixation} \mathrm{duration} \mathrm{in} \mathrm{a} \mathrm{subdiagram}}\times 100\%$$

(3)

$$\mathrm{R}4=\frac{\mathrm{Fixation} \mathrm{duration} \mathrm{in} \mathrm{the} \mathrm{incorrect} \mathrm{postures}}{\mathrm{Fixation} \mathrm{duration} \mathrm{in} \mathrm{the} \mathrm{correct} \mathrm{postures}}\times 100\%$$

(4)

$$\mathrm{R}5=\frac{\mathrm{Fixation} \mathrm{duration} \mathrm{in} \mathrm{visual} \mathrm{cues}}{\mathrm{Fixation} \mathrm{duration} \mathrm{in} \mathrm{the} \mathrm{incorrect} \mathrm{postures}}\times 100\%$$

(5)

R1 refers to the attention allocation ratio between the diagrams and texts. The fixation duration is the time spent on fixations. Fixation means when our eyes essentially stop scanning the scene, holding the central foveal vision in place so that the visual system can take in detailed information about what is being looked at. It is generally associated with attention, visual processing, and information absorption [28]. A longer fixation duration indicates that more attention is allocated. Since the area size of diagrams and texts on each stimulus is inconsistent, the per unit area value is calculated to eliminate the effects. R2 refers to the attention allocation ratio between starting positions and execution postures. R3 indicates how largely subjects distributed their attention to the key body parts in each posture by quantifying the percentage of fixation duration within the key body parts inside a whole sub-diagram. R4 and R5 are for G2 subjects. R4 computes the attention allocation ratio between the correct and incorrect posture diagrams. R5 can indicate how much subjects are attracted by the visual cues in the incorrect demonstrations.

How much effort is required to perceive the stimulus, i.e., cognitive load, is indicated by pupil diameter. Since pupil size is correlated with the level of cognitive load that the user is experiencing, pupil diameter has been a frequent indicator in reflecting varying levels of cognitive load [34]. Many previous studies have also reported that changes in cognitive load could be indicated by changes in pupil response [34]. Specifically, as cognitive load increased, pupil diameter also increased [35].

2.3. Statistical Analysis Method

The data were analyzed via Statistical Product and Service Solutions (SPSS). The two-way repeated measure ANOVA was used to compare the differences of task accuracy between the two groups of participants and to calculate the effects of instructions on learners’ learning outcomes. Mauchly’s test of sphericity was used to evaluate the normal distribution of the data at different levels. Greenhouse–Geisser’s analysis was used to correct the degree of freedom when the sphericity hypothesis was rejected. Tukey’s test for post hoc analysis was carried out to examine the difference between the means. Moreover, the two-way ANOVA was used to compare the differences in the attention allocated to various elements within the instructions between the two groups.

Hypothesis testing for two samples is used to examine the significant differences between two kinds of instructions in the ratio metrics of attention allocation. First, the following two conditions are examined to determine whether the independent t-test can be used: (1) there is no outlier in the dependent variables and (2) the dependent variables obey a normal distribution. Otherwise, the non-parametric test method will be used. The Shapiro–Wilk test was used for the normality test because the sample size is less than 50. If the p-value of the Shapiro–Wilk test is less than 0.05, the data are regarded as not normally distributed.

3. Results

3.1. Task Performance

The results of task accuracy are shown in Figure 6. Two-way repeated measure ANOVA was used to compare the accuracy in the pre-test and post-test. The average accuracies of G1 and G2 were close to each other in the pre-test without significant differences (p = 0.070). This indicates that the two samples had the same prior knowledge about working postures. The accuracy of both groups increased significantly (p < 0.001) after stimulus interferences, indicating that both kinds of stimuli helped the subjects learn the correct working postures. In addition, the accuracy of G2 was significantly higher than that of G1 (p < 0.001), indicating that the intervention effects of S2 were more significant in improving learning outcomes.

Figure 7 shows the accuracy results of two kinds of questions (about starting positions and execution postures). Two-way ANOVA was used to conduct the statistical analysis. The accuracy of G2 was also significantly higher than that of G1 in both starting positions (p < 0.001) and execution postures (p < 0.001). Finally, the comparisons between the accuracy of starting position tests and execution postures within a group showed that the accuracy of the starting positions in G1 (p < 0.001) and G2 (p < 0.001) were significantly lower than that of the execution postures. This showed that the subjects might learn knowledge about the execution postures better than the starting positions.

3.2. Attention Allocation

The statistical results of comparisons within a group and between the two groups, in terms of fixation duration per unit area of texts and diagrams, fixation duration per unit sheet of preparatory and execution postures, and fixation duration for correct and incorrect diagrams (G2 only), are shown in Figure 8, Figure 9 and Figure 10. The results were obtained through two-way ANOVA. It shows that both groups of subjects significantly allocated more attention to diagrams than to texts (p < 0.001) (Figure 8). This indicates that the diagrams attracted the subjects’ attention more. This is consistent with the subjective feedback from the subjects. All subjects reported that they allocated more attention to diagrams instead of texts. The main reasons for this were that the diagrams were more appealing (43 subjects’ selections) and that they preferred to look at the diagrams (38 subjects’ selections). Fourteen subjects claimed they did not want to read the texts because they had already understood the postures by reading the diagrams. Nevertheless, the two groups had no significant differences in the fixation duration of texts (p = 0.857) and diagrams (p = 0.090).

Comparisons between diagrams showing preparatory and execution postures (Figure 9) revealed that subjects paid significantly more attention to execution posture diagrams, both for the participants in G1 (p = 0.002) and G2 (p < 0.001). All subjects also reported that they valued the execution postures more. The main reason is that they thought execution postures were more important than preparatory postures (52 subjects’ selections), followed by wanting to learn more from execution postures (18 subjects’ selections) and regarding the preparatory postures being too easy (7 subjects’ selections).

When observing S2 stimuli, more attention was allocated to the correct diagrams than the incorrect ones (Figure 10). The majority of subjects (48) also reported spending more time looking at the correct diagrams because they observed the right ones first, and it took time to understand them. It did not take as much time to understand the incorrect diagrams. Nevertheless, no significant differences existed in the fixation duration of starting positions (p = 0.790) and execution postures (p = 0.108) between the two groups.

The comparisons between the two groups are shown in the table. The R1 results show that texts obtained less attention than diagrams. The ratio was insignificant between the two groups. Regarding R2, it also shows there was no significant difference. However, the percentage of attention to key body parts was significantly different. G2 was significantly greater than G1, indicating that subjects observing S2 stimuli were more able to attend to critical body parts. The subjective feedback was also consistent with this, with 12 subjects in G1, while 20 subjects in G2 reported noticing the key body parts. In addition, the subjects in G2 spent 13.01% fixation duration in the visual cues in the incorrect diagrams, and all of them reported being attracted by the visual cues (

Table 2. Comparisons of attention allocation between the groups.

	G1	G2	p
	Mean ± SD	Mean ± SD
R1	19.45 ± 21.47	24.14 ± 23.60	0.264 a
R2	80.11 ± 59.23	93.75 ± 91.31	0.674 a
R3	14.15 ± 5.31	20.35 ± 3.66	<0.001 a **
R4	-	10.12 ± 4.19	-
R5	-	13.01 ± 4.59	-

Note: unit (%); ** p < 0.01. ^a Abnormal distribution and the Mann–Whitney U was used.

).

3.3. Cognitive Load

Figure 11A,B indicates the values of the NASA scores and pupil diameter using boxplots, where the central mark indicates the median. The bottom and top edges of the box indicate the first and third quartiles. The highest and lowest positions indicate the maximum and minimum. According to the subjects’ NASA scores, both groups felt that it was not difficult to understand the stimulus: G1 (mean = 6.73, SD = 3.53) and G2 (mean = 6.15, SD = 2.63), and there were no significant differences in subjective cognitive load between the two groups (p = 0.060). Pupil diameter was also insignificant (p = 0.368): G1 (mean = 3.286, SD = 0.387) and G2 (mean = 3.290, SD = 0.333). This indicates that the subjects’ efforts to perceive the stimuli were not significantly different.

4. Discussion

Visual communication, such as the use of atlases, has been widely applied to different sectors with the aim of educating and informing [36]. However, few studies in the construction field have focused on using visual communication (atlases) as a support tool in occupational safety and health training (i.e., education on reducing WMSDs) among construction workers. This study compared construction workers’ attention allocation and their cognitive load when learning safe working postures from two kinds of instructional atlases with various designs.

The results of fixation duration in texts and diagrams show that all workers significantly allocated more attention to diagrams regardless of the stimulus type. This may be because the diagrams were more appealing, and the workers were prone to fixating on diagrams, according to the workers’ subjective feedback. There were even some debriefings; they did not want to read the texts because they felt they already understood the postures by reading the diagrams. The diagrams preferences are consistent with some past studies that found students were more attracted by the graphical parts of learning materials instead of the verbal parts [37,38]. Similar findings that the adoption of visual tools such as posters, pamphlets, displays, and 3D models could increase trainees’ comprehension of safety information and attract users’ attention have also been reported in various fields among the extant literature (e.g., [39,40,41]). For example, a graphic novel provides an attractive medium to communicate business concepts that may aid student learning in an effective manner [39]; a pictogram that clearly shows concrete action in which the worker and the cause of the accident are depicted can be better comprehended [42]. The findings in this study indicated that diagrams have a greater attraction to the construction workers than texts, which suggests that describing working postures with diagrams would achieve better learning results.

Regarding the comparisons between starting positions and execution postures of construction activity, the workers significantly paid more attention to these execution postures. These workers reported that starting positions were less critical to performing a task correctly, and there were fewer for them to learn than execution postures, so they tended to ignore these diagrams representing starting positions. However, incorrect preparation postures could lead to body injuries [43], and it might even affect the correctness of the following execution postures [44]. The results of task performance also showed consistently that the accuracy of questions about preparation postures was significantly smaller than that about the execution postures. Workers who do not subjectively realize the importance of the ready postures caused attention bias, so raising their awareness about the preparation postures is necessary.

In terms of attention distributed to the key parts within the instructional atlas, the attention for critical parts of workers learning from the S2 stimuli attention was significantly greater than that of workers observing S1 stimuli. When asked if they could focus on the key body parts in each figure, there was also a larger number of subjects reporting positive replies in G2 than in G1. The subjects paid more attention to the body parts involved in the force generation of a certain posture, which may be due to the help of the diagrams showing incorrect postures. The erroneous diagrams reflected commonly incorrect postures as well as their consequences (using visual cues in the possibly injured body parts). As the subjects in G2 reported, their attention was easily drawn by the visual cues. These cues were positioned at the key body parts, and visual cues have been found to attract and guide learners’ attention [45], which helped the subjects focus on the critical body parts. The results suggest that training using visual cues to emphasize the critical body parts is useful in drawing and guiding learners’ attention to the key body parts of a working posture. Similar design suggestions were also reported by the study that developed visual safety training material based on migrant farmworkers’ needs through combining graphic composition rules with a user-centred design approach [46]. It was suggested that the critical aspects should be highlighted when using cartoons and photographs showing how to perform a task.

Fixation duration was significantly larger in the correct diagrams than in the incorrect diagrams within the S2 stimuli. When asking the subjects’ allocated attention between the two, they reported that the wrong postures drew them to see it; nevertheless, they did not need to view it as long as the correct pictures did because they had already spent considerable time understanding the correct postures. However, viewing the errors was effective, guiding the subjects to key body parts. The results of task accuracy also confirmed the merits of the errors, with a significantly higher accuracy for the subjects in G2. The advantages of applying wrong demonstrations displaying consequences have also been demonstrated by previous studies. For example, a previous study [18] that investigated the comprehension of safety pictograms affixed to agricultural machinery among migrant farmworkers proposed that highlighting the actual consequences in the case of an accident could enhance pictogram comprehensibility.

Although the two differently designed atlases impacted the subjects’ attention allocation, there was no significant difference in their cognitive demands. The results of pupil diameter and the NASA survey consistently showed that there was no significant difference in the mental effort required to understand the two types of posture instructions. This indicates that both categories were easy to understand, which confirmed the effectiveness of training workers using the atlas approach. The results demonstrated the superiorities of using visual communication for educating in reducing trainees’ cognitive load.

To conclude, this study conducted a preliminary investigation on the effects of atlas design on construction workers occupational health (reducing WMSDs) education. It demonstrated that the visualization design would impact workers’ attention allocation during learning, thus impacting their learning outcomes. The findings could extend knowledge about the relations between the design of visual instruction materials and the effectiveness of construction workers’ occupational health training. Although training workers’ safe postures during construction tasks is an effective approach to reduce their WMSDs, previous studies in the construction field have failed to examine the effects of instruction design on workers’ learning outcomes so as to obtain insights in developing more effective education materials. For the construction industry, suggestions about instructional atlas design could be obtained from the findings, including applying diagrams more instead of texts to describe the instructions, emphasizing the importance of preparation posture when performing a specific construction task, and adding wrong demonstrations showing consequences with visual cues being positioned.

However, there are some limitations to the current study. First, the participants of this study are limited to novice workers with a short duration of construction experience. Future studies are suggested to consider the individual differences among workers, such as experience duration, so as to provide more targeted insights for designing appropriate instructions. Second, this study is limited to attention allocation, cognitive load, and learning outcomes. Although these factors are crucial, it failed to consider the actual reduction in workers’ WMSDs. Future studies are suggested to consider this to provide a more comprehensive evaluation of the effectiveness of occupational health training materials. Third, this study compared the subjects’ original knowledge about safe working postures with their knowledge after learning the instructional atlases, which could indicate the effectiveness of the instruction atlases against no training. Nevertheless, it failed to compare with the condition where subjects received training without the use of instruction atlases. This is suggested in further studies because it could help demonstrate the effectiveness of instruction atlases against alternative training methods.

5. Conclusions

This study compared the differences in attention allocation and cognitive load when construction workers learn from two types of instruction atlases with different designs for educating them about working postures. The results showed that the various designs led to significant differences in workers’ attention allocation and learning outcomes, but there were no significant differences in cognitive load. Specifically, the workers were more willing to fixate on diagrams rather than texts, so it is recommended to use diagrams in instructions as much as possible to attract the learners’ attention. Moreover, there was less attention paid to the preparatory postures than the execution postures, resulting in the knowledge that the preparatory postures were not well learned by the workers. It is hence recommended to emphasize the importance of preparatory postures among workers. Finally, the workers significantly allocated more attention to the key body parts with the help of error diagrams showing the consequences of wrong postures and using visual cues as warnings, which ultimately facilitated the improvement of the workers’ learning outcomes. Therefore, the use of error contrasts is recommended in instruction atlases.

For academia, the findings revealed the relations between the design of the instruction atlas teaching workers correct postures during construction tasks and the workers’ cognitive patterns (attention allocation and cognitive load), which could extend the knowledge about the relations between the design of visual instruction materials and the effectiveness of construction workers’ occupational health training. For practice, it could provide insights for designing an appropriate instruction atlas in educating workers on the right postures to perform construction tasks so as to avoid WMSDs.

Future studies are suggested to compare the effectiveness of instructional atlases with other instructional methods, such as videos or interactive training programs. This could provide a more comprehensive understanding of the effectiveness of different instructional approaches by incorporating such comparisons. In addition, it is also recommended to examine the long-term effects of instructional atlases, which could help evaluate the sustainability and effectiveness of instructional materials in reducing WMSDs among construction workers.