Technique for the Retrospective and Predictive Analysis of Cognitive Errors in Maritime Pilotage Operation

This paper presents one of the human error identification method called Technique for the Retrospective and Predictive Analysis of Cognitive Errors (TRACEr). This method has been actively used in the human reliability studies of air traffic management and the rail operations. Interestingly, there are only a few published literatures where this method has been applied to maritime operation. Therefore, this paper was attempted of applying the TRACEr methodology in order to analyze and classify the maritime pilots’ cognitive errors. For that reason, 50 accident investigation reports in the maritime pilotage operations were reviewed to demonstrate TRACEr capability. Finally, concluding remarks and further extensions in practicing TRACEr are discussed at the end of this paper.


Introduction
Technique for the Retrospective and Predictive Analysis of Cognitive Errors (TRACEr) is a comprehensive approach for human error identification and it was initially developed for the UK National Air Traffic Services following the need to analyze the aircraft proximity (Airprox) accidents reports [1]. Accordingly, this method has been applied in a number of European Organization projects for the safety study of air navigation (EUROCONTROL), particularly the study of air traffic controller errors [2,3]. TRACEr was also adapted and applied in the rail industry by the UK Rail Safety and Standards Board (RSSB). In the beginning of applying TRACEr (Rail), it has been used to analyze the accidents cause by the train driver errors [4]. Subsequently, TRACER has appeared to be the most frequently used of HEI tool for the other rail safety studies in the UK and Australia [4,5,6,] Furthermore, this tool has been used in the medical industry in order to develop medical human factor taxonomies [7]. Recently, TRACERr has been adapted for maritime operation accident investigation [8] and [9]. Therefore, it is believed that, TRACEr should also suitable for pilotage operations because of some similarities with the aforementioned studies in the context of the type of tasks, performance shaping factors and human cognitive aspect.

Literature Review
The theoretical framework of TRACEr was developed on two established scientific theories, based on Wickens model [10] and Simple Model of Cognition (SMoC) of Rasmussen [11]. Neville has described in specific the relationship between these theories [12]. Theoretically, TRACEr was developed in an interactive fashion with input from a variety of activities, including an experimental study, analysis of real accident reports, interview and discussion with experts [1]. As shown in Table  1, TRACEr comprises eight taxonomy groups which were derived from the cognitive error theories that described earlier [1,4,6]. In general, TRACEr is divided into two approaches namely retrospective and predictive. Both of these approaches use a set of a specific taxonomy and related guidewords to identify human errors. However, due to the main objective of the study in this paper which is to identify and classify the pilot errors from the accident investigation reports, therefore the first approach (retrospective) is more suitable to apply. The process of developing individual taxonomies within this new method has been conducted by the main researcher. In order to ensure the new taxonomies are valid and relevant to be applied in the pilotage operations, two experts were appointed for the reviewing process. Both experts have more than 15 years of experienced in investigation of maritime accidents and also have the background in navigation and pilotage operations. They were required to review a full set of taxonomy and classification. This set consists of taxonomies groups similar to the Table 1, but with more specific on the pilotage operations aspect. As a result, after considering the experts views, some modifications have been made including changed the task error taxonomy and the PSFs. Table 2 summaries the modifications that have been made on the previous versions of TRACEr for its application in maritime pilotage operations. This category has been removed from this study due to many overlapping in terms and created confusions to many respondents. It was suggested that this category is more relevant to TRACER (predictive) approach. Performance shaping factor Main and sub factors were added if they appeared to be missing, and removed if not relevant with maritime pilotage operation. Some specific factors to pilotage operation were added such as Social and Bridge Team Management (BTM), Machinery/propulsion and human machine interaction and Aspect of Communication and Information.
The flow diagram in Figure 1 represents the systematic process of the new TRACEr (retrospective) methodology and it is consists of 6 main steps. Each of the selected accident report will be analyzed one by one via these steps. The classification processes along with this analysis were made used of the specific maritime pilotage taxonomy groups which have been modified and validated earlier.

3.1.Background of the Case Studies
A total of 365 accident investigation reports were reviewed. All of these reports were related to maritime pilotage accidents which are available to public access in several databases provided by the accident investigation bodies and ship classification societies. However, after a number of exclusion stages, only 153 accident reports are meeting the criteria of analysis. However, for the purpose of demonstrating the application of the new TRACEr (retrospective) methodology, only 50 out of 153 accident investigation reports were analyzed as shown in Table 3. These reports were selected as all of it contained a full description on pilot cognitive errors or the descriptions on the involvement of pilot error in such accidents. The criteria for selecting the reports were as followed: a) Covered the period of accidents between 2007 and 2016. b) Accident is occurred during pilotage operation when the pilot is onboard such a case with any pilotage operation conducted by the ship's master who has held the PEC is not considered. c) Select the pilot error accidents involved with the merchant vessels more than 500 GRT thus all naval and government agency vessels are not considered.  The whole process of demonstrating the application of TRACEr (retrospective) methodology was based on the diagram flows in Figure 1. As a result, the analysis and its findings have been summarized as follow:

Step 1: Identifying of errors from the selected accident investigation report
Each of the 50 accident reports were reviewed one by one and all the pilot errors identified from each of these reports were recorded on the recording form. Overall, 106 pilot errors were identified and this represents an at least one pilot errors were occurred in each pilotage accident report

Step 2: Classify the task of each pilot error
Each of the errors identified in Step 1 then requires to be classified into the best task taxonomy which can be selected from the list of pilot task error. This list consists of nine tasks and only one of the best tasks that the error falls under should be selected. For instance, if the report mentioned that "the pilot failed to assess the effectiveness of the bridge team members" therefore, this error will be considered under the task of co-operating with the bridge members and functioning within it. The findings of the task error analysis are shown in Tables 4. It can be seen that 18 errors are classified as error in the task of co-operating with the bridge team and functioning within it. The result may be explained by the fact that this task is strongly related to the concept of the bridge team management which required the pilot to integrate fully within the bridge team. Further analysis on this task has showed that many pilots were failed to establish a close working relationship with all bridge members. Some good examples of this failure are described in the accident reports such as Van Gogh and New Caterina.
It also shown that, a number of pilots were failed to exchange and seek important information with the ship's master. This types information are including an intended port passage planning, critical stages/ points of the passage, contingencies manoeuvre and defective of navigation aids. To conclude the findings in this step, the comprehensive and systematic way of identify and recording the human task error could be performed by using TRACEr. These task errors could be used for further work such as the development of the risk analysis model such as in the hierarchical task analysis (HTA) model.

Step 3: Selection of Error Domain, Internal Error Modes (IEMs) and Psychological error Mechanisms (PEMs)
This step is representing the unique of TRACEr methodology. Only a few HEI methods could analyse the human cognitive domains as comprehensively as in this steps. Therefore, this step consists of three processes. Firstly, it is required to choose the cognitive domains that potentially apply to the error under consideration. Secondly, the most likely IEM is required for the identified domain is chosen. The single most appropriate IEM should be selected from the identified domains and recorded. Thirdly, the most possible PEMs are chosen that correspond to the same domain identified in the second process. For instance, if the IEsM is selected in the Memory domain, then the PEMs are also to be selected in the Memory domain. Conversely, more than one PEM can be selected for each error. IEMs and PEMs relates to the functions of the cognitive domain. IEMs describe the internal taxonomy of the pilot's error within each cognitive domain. While, PEMs describe how the error occurred in terms of its psychological mechanism within each cognitive domain.  Table 5 shows all the 106 pilot errors that have been classified into their respective cognitive domains.
However, this table only shows the result of the first process in this step which to determine the cognitive domains. From the result, it seems that the most common domain of the pilot cognitive error was that of decision making. Table 6 provides an example of the results from the second and third process in this step, with regard to internal and psychological errors respectively. This valuable information is almost impossible to obtain without applying any comprehensive HEI method such as TRACEr. This kind of information is very useful especially in the process of building the cognitive error model for a detail human error study. Thus, it can be suggested that the TRACEr method managed to highlight and explained the errors concerning human cognitive aspects.

Step 4: Information of error / subject of error
The information taxonomy describes the subject of error, and the terms within the taxonomy which are related to the IEM. The information taxonomy contains over 25 keywords. An example of the information taxonomy is shown in Table 7.

Step 5: Performance Shaping Factors (PSF).
In this step, the PSFs taxonomies are selected from the list of the 9 PSFs Group. Fundamentally, the PSF will identify an individual factor of the contextual human error associated with each particular error. In other words, PSF might help to explain why the pilot made the errors. Table 8 shows the groups and quantity of PSFs associated with the 106 pilot errors. In TRACER, all PSF taxonomies are in negative form in order to indicate the negative impact on the contextual human factors issues [3]. The results of this analysis indicated that the aspect of communication and information sharing was the largest impact on the performance of pilot. These findings are consistent with those of [8,13,14] who have found that the most common element of the accidents during the pilotage operations was due to inadequate master-pilot information sharing. Taken together, these results suggest that information sharing process should be continuously delivered in many stages during the operations and not only before the commencement of the pilotage operation. Other prevalent PSFs included Aspect of Communication and Information sharing, Training, Competency and Experience and Internal and External Environmental.
In conclusion of the findings in this step, TRACEr has attempted to include more systematic PSF contributing to accident occurrence and also classify the error of other personnel. The analysis of PSF was intended to avoid against the perception that human error study solely blame the operator for doing error without classify the other aspects that shaping their performance. Finally, the error should be classified into four specific terms namely causal, contributory, compounding or non-contributory. Basically, this will be strongly based on the analyst judgment. Table 9 has summarized the result of classifying error group for 106 pilot errors from 50 accident reports. It suggested that the error caused by the pilot are significant since 62 out of 106 errors are considered as the ultimate cause of the accidents (causal).

Contributory
Error that contributes to the incident and occurred in addition to the causal error but the error would probably still have occurred regardless.

20
Compounding Errors that made the situation worse and that occurred after the person realised that the situation was going to occur.

13
Non-contributory Other errors that occurred but had no bearing on the incident.

Conclusion
The main limitation associated with this study was the set of taxonomy and classification contained in the TRACEr framework may always be incomplete and may always lead to uncertainties. These uncertainties can be overcome by increasing the number of accident reports and compare the analysis result with the probabilistic assessment approach. For that reason, this issue might appropriate to addressed by considering how TRACEr and probabilistic risk assessment (PRA) method could work together to provide a more comprehensive solution as well as more consistent with the current value and practice in the human error studies. Consequently, it would be interesting, if the results of TRACEr are use as part of the human error assessment data in the form of probability values.
In conclusion, this paper has demonstrated to use TRACEr (retrospective) methodology in facilitating the process of human errors identification in maritime pilotage operations. This method has a significant contribution to increase the consistency of findings as well as to prevent the possible manipulation of data in maritime accident investigation reports. Furthermore, this method will serve as a base for future HRA studies especially on the human cognitive errors aspects. Identifying the human cognitive errors that frequently result in the occurrence of maritime accidents can lead to the development of appropriate prevention and recovery options in Human Reliability Assessment (HRA) process.