This section presents the descriptive as well as inferential statistics of the results obtained from the study. To evaluate whether the differences seen in the descriptive results reach statistical significance, analyses of variances (ANOVAs) were performed. Moreover, the Greenhouse–Geisser correction (i.e., ) or Hynh–Feldt correction (i.e., ) were applied (i.e., the significant Mauchly’s sphericity test).
5.2. Inferential Statistics
In the analysis, the within-subject factor “process model” (six levels: performance measures of RQ1–RQ3 for P1–P6) and the between-subject factor “modularization type” (three levels: flattened process models ➀, process models with groups ➁, process models with subprocesses ➂) were examined. Since not all process models (P1–P6) in the study had the same number for a modularization type, the little MCAR test was used as part of the data analysis. Based on the resulting data set, the repeated measure ANOVA was used for data analysis.
Table 5 presents the performance measure results with respect to RQ1–RQ3 for P1–P6. The main effect (ME) of performance measures for P1–P6 (ME1) and for the modularization type comparison (ME2) were evaluated, along with the interaction effect process model ∗ modularization type (IE). In addition, in the event of significance for ME1, repeated contrasts were employed. Moreover, in the event of significance for ME2, post hoc analyses using the Bonferroni post hoc criterion were employed. Finally, all statistical tests were performed two-tailed, and the significance value was set to
p < 0.05. The data for the individual process models and the data for the respective modularization types are based on the original dataset.
5.2.1. Comparison of Process Models
In
Table 5, ME1 presents the main effect for process models P1–P6 in terms of performance measures, which include Intrinsic Cognitive Load (ICL), Germane Cognitive Load (GCL), Extraneous Cognitive Load (ECL), Perceived Usefulness of Understandability (PUU), Perceived Ease of Use (PEU), Intention to Use (IU), number of fixations, and solving duration. Additionally,
Table 6 gives an overview of the performance measures of the process models that demonstrated significant results concerning the main effects.
For ICL, a significant main effect was found (p < 0.003). Specifically, P1’s ICL was significantly lower compared to P6. Similarly, P2, P4, and P5 also reported significantly lower ICL than P6. In the case of ECL, a notable main effect was observed (p < 0.001), with P1, P2, P3, P4, and P5 having significantly lower ECL compared to P6.
Regarding PUU, a substantial main effect was recorded (p < 0.001). Here, P1, P2, P3, P4, and P5 scored significantly higher than P6. The trend continued with PEU, where a significant main effect (p < 0.001) was reported, and P1, P2, P3, P4, and P5 were rated significantly higher than P6. The same was true for IU, which exhibited a significant main effect (p < 0.001), with higher scores for P1, P2, P3, P4, and P5 compared to P6.
In terms of the number of fixations, a main effect was detected (p = 0.030). Here, P5 had a lower number of fixations than P1, P2, P3, and P6. Lastly, for solving duration, there was a main effect (p = 0.005), with P5 exhibiting a shorter solving duration than P1, P2, P3, and P6.
These findings highlight the varying impacts of different process models on cognitive load, comprehensibility, and performance measures, offering critical insights into the design and efficacy of process models.
5.2.2. Consideration of Research Question RQ1–RQ3
This subsection presents the results with respect to RQ1–RQ3.
Results for RQ1
For RQ1, all three dimensions of the cognitive load theory (i.e., ICL, GCL, and ECL) were analyzed. The Huynh–Feldt correction was utilized (i.e., significant Mauchly’s sphericity test) for comparisons over all process models with each other as was higher than 0.75.
- (a)
Regarding ICL, no significant main effect can be reported while comparing the modularization types with each other over all process models. For the comparison of the single process models, a difference in the ICL can be reported for P1 (p = 0.014). ➀ (M = 0.72, SD = 0.83) and ➁ (M = 0.89, SD = 0.78) had a lower ICL than ➂ (M = 2.07, SD = 1.06). In total, as no significant main effect was shown over all process models for ICL, the null hypothesis (i.e., ➀ = ➁ = ➂) occurred. But, when comparing the ICL regarding the modularization types of the single process models, for P1, a significant difference was reported. Hence, the alternative hypotheses (i.e., ➀ < ➁ < ➂) can be stated for P1.
- (b)
Regarding GCL, no significant main effect can be reported for the comparisons over all process models. For the comparison of the single process models, no difference can be reported. In total, as no significant main effect was shown for GCL, the null hypothesis (i.e., ➀ = ➁ = ➂) occurred.
- (c)
Regarding ECL, no significant main effect can be reported while comparing the modularization types with each other over all process models. For the comparison of the single process models, no difference can be reported. In total, as no significant main effect was shown for ECL, the null hypothesis (i.e., ➀ = ➁ = ➂) occurred.
Results for RQ2
For RQ2, PUU, PEU, and IU were considered.
- (a)
Regarding PUU, no significant main effect can be reported while comparing the modularization types with one another over all process models. For the comparison of the single process models, a difference in the PUU can be reported for P5 (p = 0.016). ➁ (M = 5.91, SD = 0.94) had a higher PUU than ➂ (M = 3.93, SD = 1.65). In total, as no significant main effect was shown over all process models for PUU, the null hypothesis (i.e., ➀ = ➁ = ➂) occurred. But, when comparing the PUU regarding the modularization types of the single process models, for P5, a significant difference was reported. Hence, the alternative hypotheses (i.e., ➂ < ➀ < ➁) can be stated for P1.
- (b)
Regarding PEU, no significant main effect is given while comparing the modularization types with each other over all process models. For the comparison of the single process models, no difference can be reported. In total, as no significant main effect was shown for PEU, the null hypothesis (i.e., ➀ = ➁ = ➂) occurred.
- (c)
Regarding IU, no significant main effect is given while comparing the modularization types with each other over all process models. For the comparison of the single process models, a difference in the IU can be reported for P5 (p = 0.044). ➁ (M = 5.50, SD = 1.51) had a higher IU than ➂ (M = 3.15, SD = 2.14). In total, as no significant main effect was shown over all process models for IU, the null hypothesis (i.e., ➀ = ➁ = ➂) occurred. But, when comparing the IU regarding the modularization types of the single process models, for P5, a significant difference was reported. Hence, the alternative hypotheses (i.e., ➂ < ➀ < ➁) can be stated for P5.
Results for RQ3
For RQ3, the number of fixations such as the duration while solving the four tasks for each process model (i.e., P1–P6) were considered.
- (a)
Regarding the number of fixations, the clusters (see
Table 2 in
Section 4.3) were mapped in terms of the different modularization types (i.e., ➀, ➁, and ➂). Regarding the number of fixations, no main effect can be reported. No difference can be reported for the comparison between the modularization types. For the comparison of the single process models, a difference in terms of the number of fixations can be reported for P4 (
p = 0.035) and P6 (
p = 0.044). While comparing the process model modularization types, no significant difference was reported. In total, as no significant main effect was shown over all process models for the number of fixations, the null hypothesis (i.e.,
➀ = ➁ = ➂) occurred.
- (b)
Regarding the solving duration, no significant main effect can be reported for the modularization types over all process models. Differences can be reported while comparing single process model modularization types. A difference is given for P4 (p = 0.041). The solving duration was higher for ➂ (M = 59.49, SD = 24.63) than for ➁ (M = 37.16, SD = 14.55) and ➀ (M = 37.62, SD = 13.49). A further difference can be reported for P6 (p = 0.034). ➀ (M = 40.09, SD = 17.26) was lower than ➂ (M = 72.49, SD = 25.55). In total, as no significant main effect was shown over all process models for the solving duration, the null hypothesis (i.e., ➀ = ➁ = ➂) occurred. But, when comparing the IU regarding the modularization types of the single process models, for P4 and P6, a significant difference was reported. Hence, the alternative hypotheses (i.e., ➀ < ➁ < ➂) can be stated for P5. Furthermore, the alternative hypothesis (i.e., ➀ < ➁ < ➂) was given for P6.
In summary, when evaluating process model comprehension (addressed in RQ1 and RQ2) and performance (covered in RQ3), no significant main effect emerged from the comparison of different modularization types (➀, ➁, ➂). However, a closer examination of individual process models in terms of comprehensibility reveals supportive evidence for the use of modularization, particularly through the implementation of groups. This finding suggests that while the overall modularization type might not significantly impact comprehension and performance, specific approaches within modularization, such as grouping, can enhance the comprehensibility of process models.
In addition to the research questions RQ1–RQ3, the preference of the participants was of interest in terms of the modularization types. No significant differences can be reported regarding the preferences, F(2.00, 48.00) = 1.35, p = 0.270, and η = 0.053. Concerning the results, the following average values can be reported for the modularization types: M = 1.96 for ➀, M = 2.28 for ➁, and M = 1.84 for ➂. These results suggest that while there are variations in preferences for different modularization types, these differences are not statistically significant. This implies that participant preferences for one type of modularization over another are relatively balanced and do not strongly favor a particular style.
5.3. Comprehension Strategies
In this section, we discuss the findings related to Research Question 4 (RQ4), which involved an in-depth analysis of the data gathered from eye-tracking and audio recordings. These recordings were integrated into videos, providing a comprehensive view of participant behavior during the tasks. Each video, corresponding to a specific task within each process model, was meticulously analyzed. The focus for eye-tracking data was primarily on the sequence of eye movements. This involved documenting the order in which participants viewed different activities in the process model, ascertained through both visual eye-tracking data and corroborated by audio recordings. For instance, if a participant first looked at one activity and then immediately moved to the adjacent activity, this sequence was noted. Audio recordings were scrutinized for specific keywords that participants used while navigating through the process models (refer to
Table 7 for details). For example, “sequential” indicated a step-by-step approach through the process model. From this combined analysis of eye-tracking and audio data, various comprehension strategies employed by participants were identified. If multiple strategies were used for a single task, they were documented in the order of their occurrence. However, if the same strategy was employed consecutively for the same task, it was considered related. For example, in
Figure 5a, if a participant first focused on activity A, then B, and subsequently C, this progression was noted as a continuous strategy. After the initial documentation, these strategies were reviewed and compared in detail. It was observed that some identified strategies were composites of different strategies and were thus removed from the analysis. Subsequently, all of the videos were re-examined to ensure that the remaining comprehension strategies accurately reflected the video content. This led to a second round of documentation for each task and participant. This rigorous analysis identified and outlined seven distinct comprehension strategies, offering valuable insights into how participants interact with and comprehend process models. In the following, the seven comprehension strategies are presented:
Temporal allocation
The method for determining temporal allocation in the study involved participants associating activities with specific time segments within the process model, such as the beginning or end. This was achieved through a combination of verbal cues and eye-tracking data.
Figure 9a illustrates an example of temporal allocation. Out of the total participants, 22 successfully engaged in this temporal allocation.
This allocation process was primarily based on data obtained from the concurrent think-aloud approach, supplemented by eye movement analysis. Through the think-aloud method, participants often used keywords like “beginning” (referenced in
Table 7) to describe their focus within the process model. When the quality of eye-tracking data was sufficient, these verbal indications were cross-referenced with the participants’ eye movements, particularly to verify if their gaze shifted to areas of the process model they verbally referenced, such as the start or end of the model.
The eye movements, while beneficial for confirming the participants’ statements, were not always necessary for deducing temporal allocation as the verbal cues provided clear and direct indications of the participants’ focus areas within the process model. This approach allowed for a nuanced understanding of how participants navigated and conceptualized the process model in terms of time.
Sequential execution
In the sequential execution strategy, participants systematically progress from one activity to the next adjacent one, maintaining a consistent direction throughout. This approach includes the possibility of traversing activities from left to right or vice versa.
Figure 9b illustrates this method, labeled as erratic execution. For 23 participants, this sequential execution strategy was confirmed. This strategy was identified through a combination of data from the concurrent think-aloud approach and eye-tracking analysis.
The concurrent think-aloud protocol played a crucial role in identifying this strategy, particularly through the use of specific keywords such as “sequential” (as referenced in
Table 7). While some keywords were clear and unambiguous, others were more nuanced and required further interpretation. The determination of this strategy was strengthened by correlating these verbal cues with the participants’ eye movements, particularly observing their scan path as they moved from one activity directly to an adjacent one.
In addition to specific keywords, the verbalization of each step in a sequential order by participants also aided in identifying this strategy. The combination of these verbal indicators with the eye movement patterns provided a comprehensive understanding of how participants engaged in sequential execution while navigating through the process model.
Erratic execution
In the erratic execution strategy, participants engage with the activities in a non-sequential, unpredictable order. Instead of progressing to an adjacent activity after viewing one, they might choose to view activities that are spatially distant from each other. This approach allows for both left-to-right and right-to-left progression through the activities.
Figure 9c visually represents this erratic execution. For 23 participants, the application of an erratic execution strategy was observed. This strategy was identified through a blend of data from the concurrent think-aloud approach and analysis of eye movements.
The concurrent think-aloud approach was instrumental in this identification process, especially through the detection of specific keywords like “single activities”, as noted in
Table 7. While some of these keywords were straightforward and clear, others presented ambiguity. The final determination of this strategy was made by marrying these verbal cues with the participants’ eye movements. Notably, the scan paths demonstrated movement from one activity to another non-adjacent activity.
In addition to the keywords, participants’ verbal descriptions of engaging with single steps in a random, erratic order further solidified the identification of this strategy. This method provided a comprehensive understanding of how some participants approached the process model in a non-linear, erratic fashion.
Call to mind
In the call to mind strategy, participants first familiarized themselves with the process model and then directly jumped to a specific activity or a cluster of activities that they remembered. This approach is illustrated in
Figure 9d. Among the study participants, 15 were observed using this erratic execution strategy. The identification of this strategy primarily stemmed from the use of keywords such as “reminded”, as detailed in
Table 7. This was discerned through the concurrent think-aloud approach.
The verbal cues provided by the participants were clear and unequivocal, making the eye movement data a supplementary rather than essential component in identifying this strategy. However, when eye-tracking data was used, it showed that participants’ gaze moved directly to the activity or cluster of activities they recalled. This strategy highlights how participants leverage their memory of the process model to navigate efficiently, focusing on specific parts they remember rather than sequentially viewing each component.
Coincidence
In the coincidence strategy, participants randomly come across the activity they are searching for. This strategy is depicted in
Figure 9e. It was observed that 24 participants employed this approach, as proven by their actions during the study. The identification of the coincidence strategy was largely based on keywords such as “direct”, as listed in
Table 7. This was ascertained through the concurrent think-aloud approach.
While some keywords provided by participants were clear and straightforward, others were more ambiguous. The final determination of this strategy was made by integrating these verbal cues with the participants’ eye movement data. Notably, in this strategy, the participants’ gaze would fixate directly on the activity they were seeking without a prior systematic search.
This approach indicates a more spontaneous and less structured method of navigating through the process model, where participants happen upon the required activity by chance rather than through a deliberate search pattern.
Creating overall picture
In the creation of an overall picture strategy, participants aim to comprehend the entire process model comprehensively. This approach was observed in 21 participants. The strategy emerged from the analysis of data collected through the concurrent think-aloud approach, where participants articulated their desire to grasp the complete process.
This strategy is characterized not just by a focus on individual activities but also on other elements such as roles within the process model. Participants using this approach often verbalized terms that indicated their intention to form a holistic comprehension of the process. A notable example of such a term is “overall picture”, as mentioned in
Table 7. This keyword and similar expressions pointed to participants’ efforts to conceptualize the entire process model rather than concentrating on isolated parts or specific activities.
This strategy reflects a more comprehensive and integrated approach to process model comprehension, where participants seek to synthesize all components of the model to form a cohesive comprehension of the entire process.
Thematic allocation
In the thematic allocation strategy, participants categorize an activity within the context of a broader, generic topic. This approach is illustrated in
Figure 9f. The strategy was confirmed in 23 participants. It was identified primarily through the data gathered via the concurrent think-aloud approach. During this process, participants used specific keywords like “allocate content”, as detailed in
Table 7. These keywords were direct indicators of a thematic allocation approach, making them clear and unambiguous.
As a result, the reliance on eye movement data was minimized in this case since the verbal cues provided sufficient information to ascertain the use of this comprehension strategy. However, when eye-tracking data was considered, it showed that participants’ gaze often focused on labels or titles, such as the designation of a collapsed subprocess or group, which aligns with the thematic allocation approach.
One notable exception was observed in the thematic allocation context. A participant, while engaging with a flattened process model (➀), referred to a cluster during task resolution. Given that this modularization type does not typically include clusters, this suggests that the participant had mentally conceptualized a cluster to aid in their comprehension. This exception highlights the flexibility and diversity in how individuals process and interpret information in process models.
In addition to identifying various comprehension strategies, this study also analyzed how these strategies were utilized across different modularization types in reading process models P1–P6, as detailed in
Table 8. The table shows the frequency of each comprehension strategy’s usage across all four tasks for each process model, broken down by modularization types (flattened process models ➀, process models with groups ➁, and process models with subprocesses ➂). For instance, in P1, the “temporal allocation” strategy was used four times for ➀, five times for ➁, and four times for ➂. The comparative analysis of usage frequency reveals that each comprehension strategy was employed across all modularization types. However, a notable variation was observed in the “thematic allocation” strategy, with significantly higher usage in ➁ (28 times) and ➂ (24 times), compared to just once in ➀. Additionally, strategies like “temporal allocation”, “call to mind”, and “creating overall picture” were not consistently used in every modularization type for all process models. For example, “temporal allocation” was not used in ➂ for P3. It is also important to note that participants often combined multiple comprehension strategies when searching for activities. These combined strategies ranged from one to five, with instances where the same strategy was used multiple times, interspersed with other strategies. This led to sequences of comprehension strategies, such as 67373 (“creating overall picture” (6), “thematic allocation” (7), “erratic execution” (3), “thematic allocation” (7), and “erratic execution” (3)), appearing multiple times.
Table 9 outlines the direct predecessor (p) relationships between these strategies. For example, “temporal allocation” (1) often preceded strategies like “sequential execution” (2), “erratic execution” (3), “call to mind” (4), “coincidence” (5), and “thematic allocation” (6). However, “thematic allocation” did not precede “creating overall picture” (6), making combinations such as 12, 13, 14, 15, and 17 feasible. Interestingly, “creating overall picture” (6) was typically initiated at the start of a task, indicating that it rarely had any predecessor. This analysis of comprehension strategy usage and its sequences provides valuable insights into the cognitive processes and tactics employed by participants when interacting with different process models and modularization types.
5.4. Discussion
This study contributes to the existing body of research on process model comprehension by specifically focusing on how different types of modularization (flattened process models, process models with groups, and process models with subprocesses) affect comprehension. To explore this, four research questions (RQ1–RQ4) were formulated and investigated. These questions addressed the comprehensibility of process models both with and without modularization, encompassing ➀ (flattened process models), ➁ (process models with groups), and ➂ (process models with subprocesses).
In the analysis of individual process models (P1–P6), P6 emerged as notably more challenging to comprehend compared to the others. This was evident across several dimensions, including intrinsic and extraneous cognitive loads, perceived ease of use, perceived usefulness of understandability, intention to use, number of fixations, and solving duration, as detailed in
Table 3 in
Section 5.1. The think-aloud approach revealed that participants often verbally expressed their surprise or confusion about the placement of activities within P6. This phenomenon relates to the concept of expectation disconfirmation, which has been shown to influence behavior and experience in other research areas, such as web search [
45].
A potential explanation for the difficulties encountered with P6 might lie in the participants’ prior knowledge or lack thereof. The scenarios presented in P1–P5 (such as refueling a car) are likely to be familiar and possibly part of participants’ daily experiences. In contrast, some participants may have been unfamiliar with the lending scenario in P6. But, prior knowledge plays a significant role in process model comprehension as it can greatly reduce cognitive load. This correlation between prior knowledge and ease of understanding is supported by previous studies [
11,
46]. Therefore, the increased difficulty in comprehending P6 could be attributed to a lack of prior knowledge about the lending process among the participants.
In RQ1, we evaluated the impact of different modularization types of process models (➀, ➁, ➂) on various aspects of cognitive load (CL)—intrinsic, germane, and extraneous. This assessment occurred after participants had engaged with each process model. When comparing these modularization types, the levels of intrinsic, germane, and extraneous cognitive load ranged from low to medium (as detailed in
Table 4 in
Section 5.1). This finding aligns with [
14], which observed similar effects in simple process models with minimal complexity. As suggested by [
47], process models varying in size and complexity could exhibit different cognitive load levels. Moreover, ref. [
48] highlights the impact of information concealment in subprocesses, such as lowering abstraction levels and detailing activities within subprocesses. This can increase mental effort due to heightened abstraction and the split-attention effect. In our study, the intrinsic load reflected an average level of interactivity among process model elements. Participants’ prior knowledge about the content may have also influenced these results. Additionally, there were no apparent difficulties in forming mental models of the provided information, indicating a manageable germane cognitive load. This could be attributed to the size of the process models and the varying difficulties (e.g., content complexity, element interaction) they presented. The design of different modularization types was perceived as suitable, suggesting an appropriate level of extrinsic cognitive load. Particularly for ➂, an effective level of abstraction was employed. This is evidenced by the fact that differences in cognitive load were primarily noted in modularization types. For instance, in the subprocess model modularization (➂), the parent process was positioned similarly to ➀ and ➁, with subprocess models visually displayed below on the same page. A potential difference could arise from the concealment of subprocesses, as commonly seen in process modeling tools. Comparing modularization types revealed only one significant difference in cognitive load, specifically in the intrinsic cognitive load for P1. This discrepancy could be linked to element interactivity in ➂, which comprises multiple process models, possibly leading to the split-attention effect [
49]. Since no other significant differences in cognitive load were observed, this study suggests that modularization types do not significantly impact cognitive load. This conclusion is consistent with findings from other studies (e.g., [
8,
10,
14]), which propose that modularized process models may inherently possess intuitive comprehensibility.
In RQ2, the study focused on the comprehensibility and acceptability of various modularization types. This entailed examining Perceived Usefulness of Understandability (PUU), Perceived Ease of Use (PEU), and Intention to Use (IU). The findings, detailed in
Table 4 in
Section 5.1, indicated that both comprehensibility and intention to use were at a medium level overall. The consistently medium levels of PUU and PEU suggest that participants might have needed clarification about the comprehensibility of the different modularization types during their evaluation. This pattern aligns with the findings in [
14], where PUU and PEU were also at medium levels for various modularization representations (horizontal, vertical, and orthogonal). Another possible explanation is that the process models had only minor variations in representation across different modularization types, with the differences mainly lying in a few elements, the style of modularization, and visualization. Specifically, for process model P5, the results for PUU indicated that ➁ (process models with groups) was more comprehensible than ➂ (process models with subprocesses). Although this was the only significant difference noted for PUU, participants generally rated ➁ as the most comprehensible and acceptable modularization type, followed by ➀ (flattened process models). This preference could be attributed to the fact that in ➁, all information is visualized in one location, thereby reducing the split-attention effect that is more likely to occur in ➀ and especially in ➂. In ➂, the split-attention effect arises because the information is distributed across different subprocess models, requiring participants to search through multiple process models. To maintain content completeness in ➂, additional elements (like start and end events and duplication of roles) are necessary, which might adversely affect comprehensibility. It is important to note that while ➁ enhances comprehensibility, it does not necessarily promote the reusability of individual modules as effectively as ➂. As for IU, no significant preference for a particular modularization type was observed, suggesting that process model readers do not have a strong inclination toward any specific type. However, the interaction effect (process model* modularization type) for IU revealed a tendency for participants to favor ➁ over ➂. This preference can be linked to the higher scores in PUU and PEU, indicating that participants tend to choose the most comprehensible approach.
In RQ3, we investigated the number of fixations and the time taken to solve tasks. The findings indicated a similarity between the performance metrics for ➀ (flattened process models) and ➁ (process models with groups). However, ➂ (process models with subprocesses) required a longer solving duration and a greater number of fixations. This could be attributed to the increased number of elements and the different visual arrangements of the process models, as discussed in RQ2. With more elements to consider in ➂, participants had more points to fixate on. Additionally, the use of the concurrent think-aloud approach meant that participants were verbalizing their thoughts and performance strategies while searching for activities, which naturally extended the time taken for task completion. The variation in how participants utilized the think-aloud approach could also contribute to differences in solving duration and fixation counts. These results align with the findings from RQ1 and RQ2, particularly regarding the preferences for modularization types.
In RQ4, seven distinct comprehension strategies were identified. These strategies could be employed either individually or in combination. Each strategy was used by more than half of the participants. Strategies like “temporal allocation”, “sequential execution”, “erratic execution”, “call to mind”, “coincidence”, and “creating overall picture” were frequently observed across all modularization types. While most comprehension strategies were applicable to all types of modularization, some differences were noted, particularly with “coincidence” in ➂. This modularization type enabled participants to find activities even without considering the parent process model, thereby bypassing a chronological sequence in subprocess models.
“Thematic allocation” was only used once in ➀. When applied to flattened process models, this strategy led to the creation of thematic clusters, potentially increasing cognitive load. Furthermore, the sequencing of comprehension strategies was analyzed. Most strategies followed a successor relationship, except for “creating overall picture”, where activities were found directly without the need for subsequent steps. Within the scope of this study, it remains unclear whether a series of several comprehension strategies offers any advantage over the use of a single strategy.
5.6. Future Work
The combined eye-tracking and concurrent think-aloud study has provided valuable insights into cognitive factors such as comprehensibility, cognitive load, and procedural approaches in the context of process model comprehension. There are several promising directions for future research:
Replicating the Study in a Practical Context: Applying the study in real-world settings would be beneficial. First, process models used in practice tend to be more complex than those created for this study. Second, employees in companies may have varying levels of knowledge about process models and their representations. Third, implementing the study with actual business process management tools instead of slide presentations could offer a more realistic scenario, especially in terms of information hiding and navigating different process levels.
Testing Single Comprehension Strategies in Scientific Institutions: Investigating individual comprehension strategies can provide insights into their advantages and limitations. For example, understanding when a strategy like “temporal allocation” is most effective, possibly in scenarios where prior knowledge is available, can be valuable. This exploration could also yield practical guidance on utilizing these strategies effectively.
Exploring Combinations of Comprehension Strategies: Researching how different comprehension strategies can be combined effectively is another intriguing area. This could lead to developing guidelines for the more efficient comprehension of process models.
Examining Different Modularization Approaches: Expanding the research to include various modularization approaches, such as horizontal modularization, may uncover additional comprehension strategies and combinations, enhancing our understanding of process model comprehension.
Considering Additional Process Model Notations: Investigating other process model notations, like Event-driven Process Chains (EPCs), which have different designs and modeling elements, could reveal new comprehension strategies tailored to these notations.
Improving the Think-Aloud Approach: Prior to the study, providing participants with a list of various elements and their standard terminology could standardize responses, leading to the more accurate mapping of comprehension strategies. The consistent use of terminology by participants would aid in more precise data interpretation and strategy identification. Overall, in terms of improving the think-aloud approach, implementing all of the tips from [
20] (e.g., assurance of a warm-up training) into the methods and evaluation could have a positive impact in terms of capturing the cognitive processes.
These future research directions have the potential to deepen our understanding of how different individuals interact with and comprehend process models, contributing to more effective and user-friendly process model design and usage.