night-time offshore helicopter operations: a survey …abstract the analysis of risks per phases of...

ABSTRACTThe analysis of risks per phases of flight is fundamental for safe helicopter operations, of which night-time offshore oil- and gas-related missions form an important part. The safe execution of such missions also depends on pilots’ recent flying practice and a stable visual approach segment prior to landing. However, the poor quality of the safety data currently available prevents accurate analysis of risk on a per-phase-of-flight basis, establishment of a meaningful flying recency requirement and identification of any preferable visual approach design. To redress these problems, this paper develops a bespoke taxonomy of phases of offshore helicopter flights and uses it as the basis for a questionnaire survey on the phase-specific risk levels experienced by pilots in the night-time,

The AeronAuTicAl JournAl December 2015 Volume 119 no 1222 1475

Paper No. 4337. Manuscript received 17 January 2015 revised version received 14 July 2015 accepted 17 September 2015.This is an adapted version of a paper first presented at The 40th European Rotorcraft Forum

hosted by the Royal Aeronautical Society in Southampton, UK, September 2014.

Night-time offshore helicopter operations: a survey of risk levels per phase of flight, flying recency requirement and visual approach technique F. A. C. Nascimento [email protected]

A. Majumdar, W. Y. Ochieng and W. SchusterImperial College London The Lloyd’s Register Foundation Transport Risk Management Centre Centre for Transport Studies Department of Civil and Environmental Engineering London UK

1476 The AeronAuTicAl JournAl December 2015

perceived optimal flying recency requirement and preferred visual approach design. With the responses obtained from pilots located in seven countries, extensive statistical hypothesis testing shows that the phases involving visual scan techniques at high speed regimes are problematic, especially the visual segment of instrument approaches. Moreover, the between-night-flights time gaps required for assured flying recency were found considerably shorter than currently standardised across the industry. Finally, no preferred visual approach technique was identified. A number of important implications have been highlighted and should form the basis for future safety interventions.

NOMENCLATUREARA airborne radar approachATPL Air Transport Pilot’s LicenceCAA Civil Aviation AuthorityCAST Commercial Aviation Safety TeamCIS Commonwealth of Independent StatesCl sum of the ranks over cases (related samples Friedman’s two-way analysis of variance by ranks test)CPL Commercial Pilot’s LicenceCST chi-square testDr. Doctor of Philosophye arbitrary error accepted (sample size estimation)EU European UnionFDM flight data monitoringFET Fisher’s exact testH high kinetic energy stateHT heavy twin-turbinei rows in contingency tables (Pearson’s chi-square test)I instrument scan techniqueIATA International Air Transport AssociationICAO International Civil Aviation OrganizationIFR instrument flight rulesIH instrument scan technique – high kinetic energy stateIL instrument scan technique – low kinetic energy statej columns in contingency tables (Pearson’s chi-square test)k number of related variables (related samples Friedman’s two-way analysis of variance by ranks test)Ks kinetic energy stateKTb Kendall’s tau b non-parametric correlationsl number of ties (related samples Wilcoxon signed rank test)L low kinetic energy stateMT medium twin-turbineMW independent samples Mann-Whitney U testsn number of non-zero differences (related samples Wilcoxon signed rank test)N number of units sampledN number of cases (related samples Friedman’s two-way analysis of variance by ranks test)

nAscimenTo et al nighTTime offshore helicopTer operATions: A surVey of risk leVels... 1477

n1 sample sizes of first group compared (independent samples Mann-Whitney U test)n2 sample sizes of second group compared (independent samples Mann-Whitney U test)nj sample size of group j (independent samples t test)Nl average rank (related samples Friedman’s two-way analysis of variance by ranks test)nij observed frequencies (Pearson’s chi-square test)NTSB National Transportation Safety Board (USA)OGP International Association of Oil and Gas ProducersProf. professorR1 sum of ranks for group 1 (independent samples Mann-Whitney U test)SD standard deviationsgn(z) sign of the value of z (Kendall’s tau b non-parametric correlations)SMEs subject matter expertsSn sums of the ranks corresponding to negative differences between paired cases of variables (related samples Wilcoxon signed rank test)Sp sums of the ranks corresponding to positive differences between paired cases of variables (related samples Wilcoxon signed rank test)sp

2 pooled variance accounting for differences in per-group sample sizes (independent samples t test)St scan techniquet independent samples t test statisticT0, T1 and T2 Functions of the number of observations and tied values (Kendall’s tau b non-parametric correlations)tj number of elements in the j-th tie (related samples Wilcoxon signed rank test)U independent samples Mann-Whitney U test statisticUK United KingdomUSA United States of AmericaV visual scan techniqueVH visual scan technique – high kinetic energy stateVL visual scan technique – low kinetic energy stateZ two-tailed value of the SD of the normal distribution associated with a desired level of confidence (sample size estimation) or related samples Wilcoxon signed rank test statisticμij expected frequencies (Pearson’s chi-square test)π estimate of the highest proportion of sampling units belonging to any category in the population (sample size estimation, categorical variable)σ estimate of the population’s SD (sample size estimation, continuous variable)ΣT Kendall’s coefficient of concordance (related samples Friedman’s two-way analysis of variance by ranks test)τ Kendall’s tau b non-parametric correlation coefficientΧ2 Related samples Friedman’s two-way analysis of variance by ranks test statistic or Pearson’s chi-square test statisticX̄i Mean value of group i (independent samples t test)


1.0 INTRODUCTIONWell-discriminated phases of flight reflect analogous environmental hazards, aerodynamic loads, task complexities and pilot activities. Hence, phases of flight receive prime consideration during both the investigation of safety occurrences and the development of pilot training syllabi(1-14). However, the phases of flight uniquely performed by helicopters are still poorly discriminated(5,15) because the current taxonomies are ill-adapted from fixed-wing operations(16-21). This impacts on the quality of the safety data collected on a per-phase-of-flight basis(11,13,22), thereby precluding the identification of specific problem areas. For example, changes in pilot control strategy associated with recurrent accident scenarios, such as the switch between visual and instrument scans during helicopter decelerating approaches in degraded visual environment(14,15,23-25), are still reported inconsistently in formal investigations. This is aggravated by the persistent absence of pilot demographic analysis which could improve understanding of pilot-related events during high-risk phases of flight.

Strategies to circumvent these problems include the complete disregard to pilot demographic analysis and the use of loose or clustered phases, such as a broad approach-and-landing phase in lieu of the individualised phases of instrument approach, visual approach, hovering and landing(1,11,12,14). The associated loss of explanatory power and potential masking of important information is evident.

Despite being inaccurate, the use of clustered phases of flight is widespread in safety analysis of public transport offshore oil- and gas-related operations. These account for a significant proportion of worldwide helicopter missions, with over 20m passengers being transported annually around the world(26). Yet, helicopter accidents cause the majority of the work-related fatalities in the oil and gas sector(27-31) and accident rates are still unacceptably high(11,22,32).

In offshore helicopter operations, flight phase-specific safety interventions have concentrated on the broad clustered approach-and-landing phase (e.g., Refs 14, 23, 24, 33-36). These inter-ventions have primarily aimed to support operations in degraded visual environment, such as the night-time(37). This is associated with both significantly higher accident rates and increased odds of pilot-related and fatal accidents when compared to daytime operations(11). The night-time is also associated with rapid decay of piloting skills, a need for more frequent recurrent training than currently practiced and various procedural inconsistencies in the final visual portion of the approach, including the lack of a standard flying technique(14,23,24,38,39).

However, the current focus of safety interventions on the broad approach-and-landing clustered phase stems from safety concerns identified primarily from individual accident investigations in the North Sea during the 1980s and 1990s(37). Even though more recent statistical analysis of worldwide accidents have confirmed the higher risk of the clustered approach-and-landing phase(11,14,22), the use of clustered flight phases in all such studies prevents the identification of specific problem phases within such a broad clustered phase. Additionally, it is possible that the risks of phases that do not form the approach-and-landing clustered phase have been overlooked. Therefore, an updated assessment of the risks across more well-discriminated phases is now required globally. Moreover, since night-time offshore helicopter occurrences are associated with pilot-related issues, a thorough assessment of possible impacts of pilot demographics on risk levels is also required(11,14,22).

These assessments cannot currently be undertaken on the basis of the analysis of accident and incident reports due to low quality and limited statistical utility of the data associated with incomplete reports, untimely analysis timeframes, inappropriate amounts of data and inadequate definitions(11,13,22,40) . Similarly, a meaningful training recency requirement cannot be derived from the data available, which equally do not support the identification of any visual approach technique which might be preferable from an aircraft’s stability, hence safety, standpoint.


Given the background above, this paper pursues three aims. The first aim is to assess the risk levels associated with the phases of night-time offshore helicopter flights through a simple and yet carefully-designed questionnaire survey applied to current pilots rigorously sampled at their opera-tional environment. The survey shall support the statistical analysis of relevant pilot demographic variables against the risk levels experienced in the riskiest phases identified.

The second aim is to establish, on the basis of the experience of the pilots sampled, which intervening time will ensure that the necessary piloting skills do not decay to an unsafe level between night flights.

Finally, the third aim is to identify whether a preferred visual approach technique exists amongst the pilots sampled. In support of these aims, the objective to develop a bespoke taxonomy of phases of night-time offshore helicopter flights is also established.

The next section of the paper reviews existing taxonomies of flight phases applied to helicopter operations and the findings relevant to offshore missions. The section that follows outlines the methodology employed for the development of the taxonomy, development and application of the questionnaire survey and analysis of the results. This is followed by the presentation and discussion of the results obtained and, finally, a set of key impacts, conclusions and recommendations are outlined. These lay the foundations for the prioritisation of safety interventions on the basis of phases of flight, including with respect to pilot demographics causing concern during the riskiest phases identified. Furthermore, for the first time, a flying recency requirement and visual approach technique may be recommended on the basis of elicited pilot experience.

2.0 TAXONOMIES OF PHASES OF FLIGHTThe logic underpinning the development of current taxonomies of phases of flight is usually unclear. For example, whilst the multi-layered categories of the taxonomy jointly-developed by the Commercial Aviation Safety Team (CAST) and International Civil Aviation Organization (ICAO(41) seem to describe the spatial position of the aircraft (i.e., an air traffic controller’s point of view), the categories of the taxonomies of the ICAO/European Union (EU) working group(20) and National Transportation Safety Board (NTSB) (3) apparently focus on the tasks performed by the aircrew. Given the different underpinning logics, comparing the results of studies based on such taxonomies is often both difficult and misleading. For example, the sub-phases within phases with similar names (e.g., ‘en-route’ at Refs 20, 41 versus ‘cruise’ at Ref. 3) are not all the same.

A set of phases was developed in Ref. 42 to reflect the different tasks performed by offshore helicopters pilots, which could be detected by flight data monitoring (FDM) devices. However, the latter requirement limited the discriminatory power of the taxonomy to FDM-detectable events, with the penalty that FDM-undetectable events, such as changes in pilot scan techniques and human factors-related nuances, were ignored.

The human factors-related phases of glider flights were captured in Ref. 7 by using process charting methods and high-level mission analysis(43,44). Whilst this led to flight operations being discriminated into meaningful and independent phases, the phases described only apply to gliders and cannot be transferred to helicopter operations.

Many other taxonomies prescribe excessively broad flight phases and hence can be criticised for a lack of explanatory power and limited utility. These include the taxonomies of the International Air Transport Associations (IATA; described in Ref. 45) and UK Civil Aviation Authority (CAA(16)). The same criticism applies to numerous studies which used clustered phases to circumvent the lack of detail of their datasets (e.g., Refs 11, 13, 14, 46-48) or addressed specific research questions that did not require detailed discrimination of the phases of flight, such as Refs 49, 50.


From all said, there is a single lack of a human factors-centred taxonomy of phases of flights with applicability to risk analysis of night-time offshore helicopter operations. The methodology described in the following section addresses this, enabling the assessment of the risk levels associated with well-discriminated human factors-centred phases of night-time offshore helicopter flights, exploration of the statistical relationships with pilot demographics and the investigation of two specific issues, i.e., night-time flying recency and visual approach technique preferences. This was done through a questionnaire survey that, whilst complementing the current knowledge derived from the accident and incident analysis, is independent of them.

3.0 METHODOLOGYSafety surveys are embedded in ICAO-mandated safety management regulations and are accepted as proactive and efficient tools for anticipating conditions which may affect safety(51,52) . Surveys have the benefit of considering the views of practitioners in the industry(14,53-55) and may provide invaluable data when accidents are rare(11,22) or occurrence data is unreliable (11,13,21,22,26,56). Extensively used in aviation (e.g., Refs 57, 58), surveys have to a limited extent covered helicopter operations, e.g., emergency medical services(53,54), offshore passenger ferrying(48,59-61) and multiple mission types(62,63).

The safety survey reported in this paper was designed as a balanced cross-sectional study to compare the risk levels attributed by pilots to the different phases of flight, with all pilots invited to rate every phase considered(64,65). Additionally, a set of demographic questions were included, as well as two direct and an open-ended question. Versions of the survey were made available in three languages to support the specific strategies used to sample regions, companies and pilots. Careful piloting of the questions, their translations and sampling strategies preceded data collection. The piloting also ensured the validity of the flight phase taxonomy developed. Details of each of such steps are reported in the sections below.

3.1 Development of flight phase taxonomy

The phases of offshore helicopter flights proposed in Ref. 42 and listed in Refs 11, 12 were improved and expanded with the process charting technique(7). This was done by assessing if the phases described in Ref. 42 covered all pilot-related tasks outlined in the literature. The hierarchical task analysis of Ref. 23 highlighted the need to discriminate the approach phase into an instrument segment and a visual segment. By analogy, the takeoff and go-around manoeuvres were also broken down into visual and instrument segments.

The phases were clustered into two higher-order levels (labelled levels 2 and 3) according to the clustering strategies introduced in Refs 11, 13, 22. To account for the specific characteristics of night-time operations, categories that reflected the helicopter’s kinetic energy states and pilot’s information sampling strategies (i.e., visual versus instrument scans) were introduced. The kinetic energy state categories build upon previous investigations that considered the impacts of speed profiles on power and lift regimes and thereby, safety. In such investigations, low airspeed regimes were associated with high power demands and low translational lift. Therefore, these regimes were deemed less safe than high speed regimes, which are characterised by low engine power demands and high translational lift(3,47). However, given the human factors interest of this study, the categories included account for the impacts of speed profiles on the stability of perceptual cues (environmental and within the cockpit, e.g., instrument flight parameters) and, consequently, the crews’ ability to fly with reference to such cues. For comparable conditions (e.g., constant flight altitude), low


speeds are associated with slow variation of perceptual cues, and therefore, improved aircraft controllability. On the other hand, high speeds are associated with faster variation of perceptual cues, and therefore, more difficult control of the aircraft with reference to such cues(66). Although further categories could be created to describe the reality of operations more accurately, e.g., a medium kinetic energy state, categories were kept binary for simplicity and statistical utility. The resulting taxonomy is shown in Table 1.

Since passenger ferrying helicopters are not fitted with auto-hover or auto-land devices, there is always a need to decouple the autopilot’s upper modes at low airspeed regimes and fly the aircraft manually with reference to external visual cues(14,23,67). Therefore, instrument scan - low kinetic energy phases (i.e., IL) were not included in the taxonomy.

3.2 Questionnaire design

Following talk-through task analysis principles(14,23,24,43), the pilots were instructed to consider typical night-time offshore helicopter flights according to their experience, prior to answering the questions posed. The questionnaire was accompanied by an invitation letter which explained the purpose of the research and provided general guidance, with the aim being to capture honest and careful responses and ensure face validity(64,68). In order to focus on the risk related to suboptimal pilot performance, the participants were also instructed to consider perfectly serviceable aircraft.

3.2.1 Demographic variables

These were collected to enable the analysis of risk ratings per demographic groups and control for potential biases stemming from the subjective nature of surveys. The demographics included the region of operation, experience (i.e., rank, years as offshore helicopter pilot, flying hours in helicopters: total, instrument flight rules (IFR) and at night; and number of night deck landings),

Table 1Taxonomy of phases of flight


level of training received (i.e., category of licence held), age, gender and helicopter type flown. These variables were a subset of those mentioned in Ref. 22 and were selected for their expected relevance, based on discussions with subject matter experts during the piloting studies (explained in Section 3.3). This was important in order to keep questions to manageable numbers and avoid non-responses.

3.2.2 Rating scales

The use of rating scales in aviation is well-documented (e.g., Refs 57, 59). Semantically-anchored(68,69) 11-point rating scales(59) were employed to assess the risk levels of each level 1 phase of flight, where 0 meant ‘no risk at all’, 5 was the mid-point (i.e., ‘somewhat risky’) and 10 meant ‘extremely risky’. Based on Ref. 70, risk was defined as ‘the likelihood that an adverse event of damaging consequence will happen’. The level 1 phases were presented to participants in the chronological order of uneventful flights, with the go around phases introduced at the end as shown in Table 1. The rating scales were presented together so that the pilots could rate the phases relatively to each other as much as in absolute terms. This overcomes the limitation of Likert scales(71), where equal intervals do not translate into scores which can be proportionately related to each other(64). The phases proposed were validated by asking participants to pinpoint and rate any flight phase that could have been missed. Finally, since the results of the studies reported in Refs 48, 59 indicated satisfactory levels of construct, predictive and concurrent validity of the rating scales when applied to offshore helicopter pilots, these studies were useful frames of reference for the present study and no further validation of the scales was attempted(43,64).

3.2.3 Direct and open-ended questions

The first direct question addressed the worldwide problem of pilot recency in night-time offshore helicopter operations(14). At present, recency requirements tend to follow the recommendations of the International Association of Oil and Gas Producers (OGP), whereby pilots are required to perform three night-time approaches and takeoffs within a period of 90 days(72). This arbitrary time gap is a legacy requirement established in the 1980s for the helicopter types of the time and may well be insufficient to ensure pilot recency at present(14,23,24,73). Pilots were asked to state in how many days after a night-time flight they felt a second night-time flight was needed so they would still feel current.

The second direct question aimed to explore pilots’ preferences for visual approach procedure geometry, i.e., levelled or sloping visual segments, typical of airborne radar approach (ARA) procedures or spatial gates, respectively(23,24). The safety of each technique has been debated inconclusively in the industry(39).

The final open-ended question had multiple aims and was used to capture any missing themes, gather participants’ suggestions on potential safety improvements and last but not least to build rapport.

3.3 Piloting

The phases of flight, scales of risk experienced, demographic, direct and open-ended questions were subject to peer-review, piloting and refinement by six subject matter experts (SMEs) based in three continents. These SMEs were five senior offshore helicopter pilots involved in training and safety within their organisations and a safety analyst who held of a master’s level degree in aviation human factors and worked for a major offshore helicopter operator. This process led to changes in the ordering, wording and content of questions for completeness, brevity and ease of response. It


increased the content, construct and face validity of the questionnaire’s questions(64). Developed and piloted in English, the questionnaire was then translated into Spanish and Portuguese. The translations were verified by native speakers who had doctoral-level education in aviation safety and human factors, as well as practical experience in such fields.

3.4 Region, company and pilot sampling strategies

Since the night-time is a high risk factor for worldwide offshore helicopter operations(11), the survey pursued worldwide coverage using the areas outlined in Refs 11, 14, 22 as the basis for a regional sampling strategy. Group 1 was formed by Brazil, California, Middle East, India, Far East, Australia, Africa and Europe except the North Sea (e.g., Italy and Spain), representing areas of typically benign weather conditions served predominantly by medium-twin turbine helicopters. Furthermore, in these areas offshore helicopter operations are not conducted routinely at night, which could affect local standard operating procedures and pilot perception of risk(14,24). Group 2 was formed by Alaska and Canada, where the weather conditions are typically adverse. Group 3 was formed by the countries of the Commonwealth of Independent States (CIS), which differ from the previous group because, in the CIS, the aircraft predominantly used are manufactured by companies from the countries of the Former Soviet Union, with potentially different airwor-thiness requirements(1). Group 4 was formed by the Gulf of Mexico and Group 5 by the North Sea. The former represents an area dominated by single engine operations and the latter is an area of tightly-regulated operations.

Within each region, the major companies operating helicopters in support of offshore oil- and gas-related missions were identified and targeted. This strategy has already been successfully applied in the offshore oil and gas industry, where very few large companies usually dominate representative market shares(28). In order to maximise the chances of reaching the relevant companies (and thereby respondents), aviation regulators, accident investigation boards, oil and gas companies and regional industry safety groups (e.g., Ref. 74) were also contacted.

The selection of representative companies was followed by a quota sampling strategy at company level, aimed at capturing the experiences of two groups: senior and flight line-only pilots. The former were pilots who dealt with other pilots in a wide range of flying conditions and had to display declarative knowledge(75) as part of the job. Flight instructors, aviation safety officer, operations and management post-holders (e.g., chief pilots and chief operating officers) were considered senior pilots. The experience of senior pilots was anticipated to most likely synthesise that of many pilots(76) and be particularly useful if sample sizes were small (e.g., if questionnaire response rate was low). The flight line-only pilot group was formed by line pilots without an administrative job and was used to control for potential professional bias of the previous group. All pilots who participated in the survey were volunteers.

The questionnaire, together with the invitation letter, was distributed by electronic mail addressed to the persons in charge of the safety, operations and management departments of the targeted companies. These persons had direct control of the sampling frame and were instructed to distribute the questionnaire according to the sampling strategy mentioned above. The language version used was the most appropriate in each case. The date of return was stipulated to be approximately two months from the date of distribution. Reminders were sent periodically to increase response rates. In four specific countries mentioned in the results section, the questionnaire was additionally applied by the main author in individual and confidential face-to-face questionnaire application sessions. Given the simplicity of the questionnaire, any potential bias introduced by face-to-face contact (e.g., social desirability) was deemed minimal and irrelevant(64).


3.5 Sample sizes

The required sample sizes were estimated as in Equations (1) and (2) below (64):

. . . (1)

. . . (2)

where σ is an estimate of the population’s standard deviation (SD), Z is the two-tailed value of the standardised deviation of the normal distribution associated with the desired level of confidence, e is the arbitrary error accepted and π is an estimate of the highest proportion of sampling units belonging to any of the categories in the population.

Three variables were used to estimate the required sample sizes: (i) the number of night-time deck landings, (ii) risk ratings and (iii) membership to the ‘senior pilot’ group. The number of night deck landings was important because the flight segment is the best variable based on which pilot-related events should be normalised(11). Risk ratings were used because they represent the main output of the investigation. Finally, membership to the senior pilot group was used because this is at the centre of the research problem.

With no prior study conducted amongst the target population (i.e., worldwide offshore helicopter pilots), σ was estimated by dividing the assumed range of the continuous variables by six. This is based on an anticipation of normality, where the spread of scores is around six SDs. Such an assumption is acceptable in the virtual absence of information for sample size estimation(64).

The range of the number of night deck landings was estimated from the minimum of three required by the OGP for night-time recency(72) to a maximum of 1,080. This considers: (i) maximum allowable of 900 flying hours per pilot per year (e.g. Refs 77, 78); (ii) 3% of flying hours at night(11,46); (iii) one deck landing per hour; and (iv) a pilot service life of 40 years. The range of risk ratings was estimated from the 0-10 scales used.

Z was initially set at 1.96, corresponding to 95% confidence interval. A function of the desired precision, e was set at 50 night deck landings, 0.5 risk rating (in order to avoid confusion between adjacent ratings) and 10% of the true proportion of senior pilots in the population. Finally, π was estimated at 0.9, assuming that typically 10% of the pilots in any company are senior pilots.

Based on these assumptions and formulae, the following per group sample sizes were calculated: (i) 50 based on number of night deck landings; (ii) 51 based on risk ratings; (iii) 35 based on seniority membership. Hence, responses were sought from at least 51 pilots in each group, with as much representativeness of the area groups outlined in Section 4.4 as possible.

3.6 Statistical analysis

Given that the risk ratings of each phase were attributed by comparison to one another, the ratings were treated as continuous data(79). Using statistical procedures that fitted the characteristics and distributions the data, the following hypotheses were initially tested:

l H01: the distributions of risk ratings across phases of flight (level 1) are the same;l H02: the distributions of risk ratings across phases of flight (level 2) are the same;l H03: the distributions of risk ratings across phases of flight (level 3) are the same;l H04: the distributions of risk ratings across combined scan technique – kinetic state (St-Ks)

categories are the same.

9

companies. These persons had direct control of the sampling frame and were instructed to distribute the questionnaire according to the sampling strategy mentioned above. The language version used was the most appropriate in each case. The date of return was stipulated to be approximately two months from the date of distribution. Reminders were sent periodically to increase response rates. In four specific countries mentioned in the results section, the questionnaire was additionally applied by the main author in individual and confidential face-to-face questionnaire application sessions. Given the simplicity of the questionnaire, any potential bias introduced by face-to-face contact (e.g., social desirability) was deemed minimal and irrelevant [64].

3.5 Sample sizes

The required sample sizes were estimated as in equations 1 and 2 below [64]:

σ (1)

(2)


Three variables were used to estimate the required sample sizes: i) the number of nighttime deck landings, ii) risk ratings and iii) membership to the ‘senior pilot’ group. The number of night deck landings was important because the flight segment is the best variable based on which pilot-related events should be normalised [11]. Risk ratings were used because they represent the main output of the investigation. Finally, membership to the senior pilot group was used because this is at the centre of the research problem.

With no prior study conducted amongst the target population (i.e., worldwide offshore helicopter pilots), σ was estimated by dividing the assumed range of the continuous variables by six. This is based on an anticipation of normality, where the spread of scores is around six SDs. Such an assumption is acceptable in the virtual absence of information for sample size estimation [64].

The range of the number of night deck landings was estimated from the minimum of three required by the OGP for nighttime recency [72] to a maximum of 1080. This considers: i) maximum allowable of 900 flying hours per pilot per year (e.g., [77, 78]); ii) 3% of flying hours at night [11, 46]; iii) one deck landing per hour; and iv) a pilot service life of 40 years. The range of risk ratings was estimated from the 0-10 scales used.


Based on these assumptions and formulae, the following per group sample sizes were calculated: i) 50 based on number of night deck landings; ii) 51 based on risk ratings; iii) 35 based on seniority membership. Hence, responses were sought from at least 51 pilots in each group, with as much representativeness of the area groups outlined in section 4.4 as possible.

9

companies. These persons had direct control of the sampling frame and were instructed to distribute the questionnaire according to the sampling strategy mentioned above. The language version used was the most appropriate in each case. The date of return was stipulated to be approximately two months from the date of distribution. Reminders were sent periodically to increase response rates. In four specific countries mentioned in the results section, the questionnaire was additionally applied by the main author in individual and confidential face-to-face questionnaire application sessions. Given the simplicity of the questionnaire, any potential bias introduced by face-to-face contact (e.g., social desirability) was deemed minimal and irrelevant [64].

3.5 Sample sizes

The required sample sizes were estimated as in equations 1 and 2 below [64]:

σ (1)

(2)


Three variables were used to estimate the required sample sizes: i) the number of nighttime deck landings, ii) risk ratings and iii) membership to the ‘senior pilot’ group. The number of night deck landings was important because the flight segment is the best variable based on which pilot-related events should be normalised [11]. Risk ratings were used because they represent the main output of the investigation. Finally, membership to the senior pilot group was used because this is at the centre of the research problem.

With no prior study conducted amongst the target population (i.e., worldwide offshore helicopter pilots), σ was estimated by dividing the assumed range of the continuous variables by six. This is based on an anticipation of normality, where the spread of scores is around six SDs. Such an assumption is acceptable in the virtual absence of information for sample size estimation [64].

The range of the number of night deck landings was estimated from the minimum of three required by the OGP for nighttime recency [72] to a maximum of 1080. This considers: i) maximum allowable of 900 flying hours per pilot per year (e.g., [77, 78]); ii) 3% of flying hours at night [11, 46]; iii) one deck landing per hour; and iv) a pilot service life of 40 years. The range of risk ratings was estimated from the 0-10 scales used.


Based on these assumptions and formulae, the following per group sample sizes were calculated: i) 50 based on number of night deck landings; ii) 51 based on risk ratings; iii) 35 based on seniority membership. Hence, responses were sought from at least 51 pilots in each group, with as much representativeness of the area groups outlined in section 4.4 as possible.


These procedures aimed to highlight which were the most critical phases of flight, according to the experience of the sample investigated. The tests of H02 and H03 additionally indicated the extent to which clustering flight phases could produce misleading results. Building on previous investigations of offshore helicopter accidents, which failed to find associations between regions of operations and phases of flight(11,22), the hypotheses tested disregarded regional differences.

With the results of the previous tests, the following hypotheses were subsequently tested:

l H05: the distributions of risk ratings across the level 1 phases forming the most critical level 3 phase (identified from the test of H03 above) are the same;

l H06: the distributions of risk ratings across the level 1 phases forming the most critical St-Ks category (identified from the test of H04 above) are the same.

Such tests confronted the test of H01 and enabled the selection of the most critical phase for the subsequent demographic analysis that aimed to identify regions or groups of pilots particularly vulnerable to the most critical phases of flight. The tests were undertaken by analysing the distri-butions of risk ratings per demographic variable collected.

Finally, the time gaps for assured recency (and conversely for recency decay) and the preferred visual approach procedural techniques were analysed against the demographic variables. The hypotheses tested and the statistical tests applied are shown in the results and discussion section. The hypotheses were two-tailed and significance was established at p ≤ 0.05.

4.0 RESULTS AND DISCUSSION

4.1 Demographics

A total of 61 questionnaires were returned, of which 40 were obtained from face-to-face sessions in Brazil, Spain, Norway and Holland. Respondents were based in Groups 1 and 5, as shown in Table 2.

There were three female and 58 male pilots, 46 captains and 13 first officers. Fifty respondents held Air Transport Pilot’s licences (ATPL) and nine held Commercial Pilot’s Licences (CPL); 23 were ‘flight line-only pilots’ and 35 respondents were ‘senior pilots’. The former corresponds to Zs that are representative of over 81.3% confidence intervals for both the number of night deck

Table 2Areas and number of respondents


landings and risk ratings. In addition, since the sample of flight line-only pilots roughly matched the size of the senior pilot group, it was deemed adequate for statistical comparisons(64). The sums that did not reach the total of 61 stem from slightly incomplete questionnaires returned online, for which follow-up was attempted in vain. Incomplete fields were excluded from the statistical analysis undertaken.

The main demographic characteristics of the pilots sampled is summarised in Table 3. This shows that pilot experience was variable and, with respect to night operations, relatively limited. This overall limited night-time experience could stem from the regional sampling strategy described in Section 3.4, whereby pilots of areas not routinely served by offshore helicopters at night were sampled. However, night-time offshore flying experience can also be limited in regions where these occur more regularly. For example, it is estimated that 8.5% of the hours flown annually in the British sector of the North Sea correspond to night operations (13). Nevertheless, regardless of the sample’s overall night-time flying experience level, any significant regional and specific experience effects should be identifiable by extensive statistical testing. This is presented in Section 4.3. The pilots flew primarily the helicopter types of Table 4.

Helicopter types were further categorised according to the following criteria:

l Transport capacity(11): heavy twin-turbine (HT, i.e., EC332L, EC332L2, EC225 and S92; N = 20) and medium twin-turbine (MT, i.e., all else; N = 37) helicopters.l Cockpit generation(13,14): full glass cockpit (i.e., S76C++, EC155, AW139, EC225 and S92;

N = 47) and analogic/mixed cockpit (i.e., all else; N = 10).

The latter enabled testing of the doubtful and yet widespread belief that glass cockpits are inher-ently safer than analogic and mixed designs(38,80).

4.2 Descriptive statistics

The medians, means and standard deviations of the ratings assigned to the flight phases can be seen in Table 5. Medians are presented since they are more meaningful when rank-based non-parametric tests are required.

Table 3Main demographic characteristics of respondents

Table 4Aircraft flown by respondents (ordered by empty weight)


Regarding the recency-related direct question, the mean time gap for recency decay reported was 28 days (SD = 56.9). Finally, in response to the second direct question, 30 and 19 pilots preferred respectively the sloping and the levelled visual approach segments.

4.3 Hypothesis testing

During all subsequent tests, the normality of the distributions were tested according to the sample sizes involved, i.e., Shapiro-Wilk’s test were used when the cases were fewer than 50 per category and Kolmogorov-Smirnov tests were used otherwise. Homogeneity of variance was checked by Levene’s test(81).

4.3.1 Preliminary statistical analysis (i.e., the tests of H01-4)

Because in all cases at least one assumption of parametric data was violated, non-parametric techniques were required to test H01-4. Related samples Friedman’s two-way analysis of variance by ranks were used to account for the fact that each respondent scored every phase of flight proposed(81).

This procedure tests if k related variables were drawn from the same population. The test sorts and ranks the k variables on each N case, with the average rank being assigned in case of ties. For each k variable, the sum of the ranks over the cases (Cl) and the average rank (Nl = Cl/N) are calculated. The test statistic is given by Equation (3):

. . . (3)

where ∑T is the Kendall’s coefficient of concordance and the significance level is derived from a Χ2 distribution with k-1 degrees of freedom(82).

The results of the tests of all four null hypotheses (i.e., H01-4) were significant and hence the hypotheses were rejected.

4.3.2 Post hoc analysis of preliminary statistical tests

Post hoc analysis was undertaken by application of the related samples Wilcoxon signed rank test with Bonferroni correction. This is used to test whether two related variables were drawn from the same population.

Table 5Medians, means and standard deviations (SD) of

flight phase ratings (highest medians in bold font)

13

Regarding the recency-related direct question, the mean time gap for recency decay reported was 28 days (SD=56.9). Finally, in response to the second direct question, 30 and 19 pilots preferred respectively the sloping and the levelled visual approach segments.


During all subsequent tests, the normality of the distributions were tested according to the sample sizes involved, i.e., Shapiro-Wilk’s test were used when the cases were fewer than 50 per category and Kolmogorov-Smirnov tests were used otherwise. Homogeneity of variance was checked by Levene’s test [81].


Because in all cases at least one assumption of parametric data was violated, non-parametric techniques were required to test H01-4. Related Samples Friedman’s Two-Way Analysis of Variance by Ranks were used to account for the fact that each respondent scored every phase of flight proposed [81].

This procedure tests if k related variables were drawn from the same population. The test sorts and ranks the k variables on each N case, with the average rank being assigned in case of ties. For each kvariable, the sum of the ranks over the cases (Cl) and the average rank (Nl = Cl/N) are calculated. The test statistic is given by equation 3:

�� ∑ �� ∑�� (3)

where ∑� is the Kendall’s coefficient of concordance and the significance level is derived from a Χ2 distribution with k-1 degrees of freedom [82].



Post hoc analysis was undertaken by application of the Related Samples Wilcoxon Signed Rank Test with Bonferroni Correction. This is used to test whether two related variables were drawn from the same population.

The test calculates the differences between paired cases of variables and all nonzero results are sorted into ascending order and ranked, with the average rank being assigned in case of ties. The sums of the ranks corresponding to positive and negative differences are calculated (Sp and Sn, respectively) as well as the average positive and negative ranks. The test statistic is as follows in equation 4.

� � min�� 4 �

� �� 4 � ∑ �� 4��

(4)

where n is the number of nonzero differences, l is the number of ties and tj is the number of elements in the j-th tie, j=1, ..., l. For large-enough sample sizes, the distribution of Z is approximately standard normal and the two-tailed probabilities are derived accordingly [83].


The test calculates the differences between paired cases of variables and all nonzero results are sorted into ascending order and ranked, with the average rank being assigned in case of ties. The sums of the ranks corresponding to positive and negative differences are calculated (Sp and Sn, respectively) as well as the average positive and negative ranks. The test statistic is as follows in Equation 4.

. . . (4)

where n is the number of nonzero differences, l is the number of ties and tj is the number of elements in the j-th tie, j = 1, ..., l. For large-enough sample sizes, the distribution of Z is approximately standard normal and the two-tailed probabilities are derived accordingly(83).

The Bonferroni correction avoids falsely rejecting the null hypothesis (i.e., committing a Type I error) by using a more stringent rejection criterion given by the initial significance level divided by the number of pairs under analysis. In each such test, the category with highest median was chosen as the control group(11,22). This enabled groups to be assessed against the group with potentially the highest risk ratings. Therefore, the relevant results are the statistically insignificant results because they indicate phases in which risk ratings are comparable to the worst of the phases.

The tests following the rejection of H01 led to 14 pairs being formed for comparison. The only ratings that failed to achieve significance (p > 0.05/14), and hence were comparable to those attributed to the ‘visual segment of the approach’ (i.e., control group), were those of the ‘visual segment of the takeoff’, ‘go around in the instrument segment of the approach’ and ‘go around in the visual segment of the approach’.

During the post hoc tests of H02, five pairs of phases were compared. The only ratings that failed to achieve significance (p > 0.05/5), and hence were comparable those attributed to the ‘landing’ level 2 phase (i.e., control group), were those of the ‘taxiing’, ‘take-off’ and ‘approach’ phases. This highlights two misleading results stemming from the clustering of flight phases, which have major implications for safety analysis of helicopter operations: firstly, ‘landing’ appeared as the most critical level 2 phase (i.e., highest median). Secondly, ‘taxiing’ figured as a level 2 phase of concern even though no level 1 categories within it achieved significance during the previous tests. Given that the level 2 phases proposed are fairly similar to the phases of widely-used taxonomies, e.g., that of the UK CAA(11,16,22), care should be taken when interpreting the results of official studies.

The post hoc tests of H03 were undertaken as three paired-tests. Using the ‘arrival’ level 3 phase as the control group, the ‘departure’ phase failed to show significant differences in pilot ratings (p > 0.05/3). Therefore, the risk levels assigned were comparable between the ‘arrival’ and ‘departure’ level 3 phases.

The post hoc tests of H04 compared the ratings of the VL and IH phases against those of the VH phase (i.e., the control group). Both tests were significant, and therefore the risk ratings were significantly higher for the VH phases.

4.3.3 Further statistical testing (i.e., the tests of H05-6)

Given the results of the test mentioned above, H05 and H06 tested the distributions of risk ratings across the level 1 phases within the ‘arrival’ (level 3) and ‘VH’ phases, respectively. The statistical tests mentioned above were applied because of similar data characteristics and sample sizes.

The test of H05 produced a significant result. The post hoc tests revealed that high ratings in the ‘visual segment of the approach’ (i.e., control group) were only comparable to those of the

13

Regarding the recency-related direct question, the mean time gap for recency decay reported was 28 days (SD=56.9). Finally, in response to the second direct question, 30 and 19 pilots preferred respectively the sloping and the levelled visual approach segments.


During all subsequent tests, the normality of the distributions were tested according to the sample sizes involved, i.e., Shapiro-Wilk’s test were used when the cases were fewer than 50 per category and Kolmogorov-Smirnov tests were used otherwise. Homogeneity of variance was checked by Levene’s test [81].


Because in all cases at least one assumption of parametric data was violated, non-parametric techniques were required to test H01-4. Related Samples Friedman’s Two-Way Analysis of Variance by Ranks were used to account for the fact that each respondent scored every phase of flight proposed [81].

This procedure tests if k related variables were drawn from the same population. The test sorts and ranks the k variables on each N case, with the average rank being assigned in case of ties. For each kvariable, the sum of the ranks over the cases (Cl) and the average rank (Nl = Cl/N) are calculated. The test statistic is given by equation 3:

�� ∑ �� ∑�� (3)

where ∑� is the Kendall’s coefficient of concordance and the significance level is derived from a Χ2 distribution with k-1 degrees of freedom [82].



Post hoc analysis was undertaken by application of the Related Samples Wilcoxon Signed Rank Test with Bonferroni Correction. This is used to test whether two related variables were drawn from the same population.

The test calculates the differences between paired cases of variables and all nonzero results are sorted into ascending order and ranked, with the average rank being assigned in case of ties. The sums of the ranks corresponding to positive and negative differences are calculated (Sp and Sn, respectively) as well as the average positive and negative ranks. The test statistic is as follows in equation 4.

� � min�� 4 �

� �� 4 � ∑ �� 4��

(4)

where n is the number of nonzero differences, l is the number of ties and tj is the number of elements in the j-th tie, j=1, ..., l. For large-enough sample sizes, the distribution of Z is approximately standard normal and the two-tailed probabilities are derived accordingly [83].


‘go around in the visual segment of the approach’ (i.e., p > 0.05/5). This challenges the post hoc tests of H01, which indicated that the ‘go around in the instrument segment of the approach’ also belonged to this population. However, the latter inclusion appears to have been a Type II error during the post hoc tests of H01, stemming from the excessively stringent rejection criterion caused by the application of the Bonferroni Correction to all 14 pairs of phases. This makes operational sense as phases imposing akin perceptual demands on pilots (i.e., visual sampling at speed in impoverished visual conditions) were rated similarly high; and the phase flown on instruments received significantly lower ratings.

Given that the test of H06 failed to produce significant results, and hence the risk ratings assigned to all VH phases were similar, the phases rated highest for risk experienced were those flown by reference to external visual cues at high kinetic energy states (i.e., the visual segments of takeoff, approach and go around). As with the test of H05 described above, this makes operational sense for the similitude of pilot perceptual demands involved.

Considering the results above, the worldwide concentration of offshore helicopter pilot-related accidents in the arrival segment of flight(11,22) and the numerous international efforts to redress this problem, the ‘visual segment of the instrument approach’ can be considered to be the single most critical phase of night-time offshore helicopter flights. Therefore, the statistical analysis that follows focuses on both this phase and the VH category.

4.3.4 Risk ratings per demographic variables

Given that in all cases at least one assumption of parametric data was violated, non-parametric tests were needed. The independent samples Mann-Whitney U test was used to compare the distributions of risk ratings across categories of nominal variables, whereas Kendall’s tau b non-parametric correlations were used against continuous variables.

In the former test, the observations in both groups are combined and ranked. The number of times that a score from one group precedes a score from the other group, and vice versa, is calculated. The Mann-Whitney U statistic is the smaller of these two numbers, given by the Equation (5):

. . . (5)

where n1 and n2 are the sample sizes of both groups being compared, R1 is the sum of ranks for group 1 (assumed with the smallest U), and the significance levels of U are given by specific algorithms(11,81,84,85).

Kendall’s Tau b measures the association between variables based on the rank correlation, i.e., the number of concordances and discordances of paired observations when the data is ranked by each of the quantities. The formula of Kendall’s tau b is shown as Equation (6):

. . . (6)

where T0, T1 and T2 are functions of the number of observations and tied values and sgn(z) equals to −1, zero or +1, respectively for negative, null or positive values of z. The test probabilities are derived from the acceptance that the quotient formed by τ’s numerator over τ’s numerator’s variance comes from a standard normal distribution(86).

Table 6 presents the hypotheses tested and the results obtained. The significance on the tests of H08 and H09 occurred as the risk ratings of VH phases were higher for medium twin-turbine and glass cockpit helicopters, respectively. Given that these results were insignificant for the ‘visual

15

reference to external visual cues at high kinetic energy states (i.e., the visual segments of takeoff, approach and go around). As with the test of H05 described above, this makes operational sense for the similitude of pilot perceptual demands involved.

Considering the results above, the worldwide concentration of offshore helicopter pilot-related accidents in the arrival segment of flight [11, 22] and the numerous international efforts to redress this problem, the ‘visual segment of the instrument approach’ can be considered to be the single most critical phase of nighttime offshore helicopter flights. Therefore, the statistical analysis that follows focuses on both this phase and the VH category.


Given that in all cases at least one assumption of parametric data was violated, non-parametric tests were needed. The Independent Samples Mann-Whitney U Test was used to compare the distributions of risk ratings across categories of nominal variables, whereas Kendall’s Tau b non-parametric correlations were used against continuous variables.

In the former test, the observations in both groups are combined and ranked. The number of times that a score from one group precedes a score from the other group, and vice versa, is calculated. The Mann-Whitney U statistic is the smaller of these two numbers, given by the equation 5:

� � �� 2 � �� (5)

where n1 and n2 are the sample sizes of both groups being compared, R1 is the sum of ranks for group 1 (assumed with the smallest U), and the significance levels of U are given by specific algorithms [11, 81, 84, 85].

Kendall’s Tau b measures the association between variables based on the rank correlation, i.e., the number of concordances and discordances of paired observations when the data is ranked by each of the quantities. The formula of Kendall’s tau b is shown as equation 6:

τ ��∑ �� T� � T�� T� � T��

(6)

where T0, T1 and T2 are functions of the number of observations and tied values and sgn(z) equals to -1, zero or +1, respectively for negative, null or positive values of z. The test probabilities are derived from the acceptance that the quotient formed by τ’s numerator over τ’s numerator’s variance comes from a standard normal distribution [86].

Table 6 presents the hypotheses tested and the results obtained. The significance on the tests of H08 and H09 occurred as the risk ratings of VH phases were higher for medium twin-turbine and glass cockpit helicopters, respectively. Given that these results were insignificant for the ‘visual segment of the approach’, the significance stems from the visual segments of either the takeoff or go around. This represents an important operational impact as pilots may find flying visual segments of takeoff or go around especially challenging when medium twin-turbine or glass cockpit helicopters are employed.

15

reference to external visual cues at high kinetic energy states (i.e., the visual segments of takeoff, approach and go around). As with the test of H05 described above, this makes operational sense for the similitude of pilot perceptual demands involved.

Considering the results above, the worldwide concentration of offshore helicopter pilot-related accidents in the arrival segment of flight [11, 22] and the numerous international efforts to redress this problem, the ‘visual segment of the instrument approach’ can be considered to be the single most critical phase of nighttime offshore helicopter flights. Therefore, the statistical analysis that follows focuses on both this phase and the VH category.


Given that in all cases at least one assumption of parametric data was violated, non-parametric tests were needed. The Independent Samples Mann-Whitney U Test was used to compare the distributions of risk ratings across categories of nominal variables, whereas Kendall’s Tau b non-parametric correlations were used against continuous variables.

In the former test, the observations in both groups are combined and ranked. The number of times that a score from one group precedes a score from the other group, and vice versa, is calculated. The Mann-Whitney U statistic is the smaller of these two numbers, given by the equation 5:

� � �� 2 � �� (5)

where n1 and n2 are the sample sizes of both groups being compared, R1 is the sum of ranks for group 1 (assumed with the smallest U), and the significance levels of U are given by specific algorithms [11, 81, 84, 85].

Kendall’s Tau b measures the association between variables based on the rank correlation, i.e., the number of concordances and discordances of paired observations when the data is ranked by each of the quantities. The formula of Kendall’s tau b is shown as equation 6:

τ ��∑ �� T� � T�� T� � T��

(6)

where T0, T1 and T2 are functions of the number of observations and tied values and sgn(z) equals to -1, zero or +1, respectively for negative, null or positive values of z. The test probabilities are derived from the acceptance that the quotient formed by τ’s numerator over τ’s numerator’s variance comes from a standard normal distribution [86].

Table 6 presents the hypotheses tested and the results obtained. The significance on the tests of H08 and H09 occurred as the risk ratings of VH phases were higher for medium twin-turbine and glass cockpit helicopters, respectively. Given that these results were insignificant for the ‘visual segment of the approach’, the significance stems from the visual segments of either the takeoff or go around. This represents an important operational impact as pilots may find flying visual segments of takeoff or go around especially challenging when medium twin-turbine or glass cockpit helicopters are employed.


segment of the approach’, the significance stems from the visual segments of either the takeoff or go around. This represents an important operational impact as pilots may find flying visual segments of takeoff or go around especially challenging when medium twin-turbine or glass cockpit helicopters are employed.

Table 6Statistical analysis of risk ratings at the most

critical flight phases across demographic variables

Table 7Statistical analysis of gap for recency decay across demographic variables


4.3.5 Time gap for recency decay per demographic variables

Given the similar data characteristics and sample sizes involved, the statistical tests mentioned above were applied. The hypotheses tested with their results are summarised in Table 7.

The rejection of H023 stemmed from shorter time gaps for recency decay reported by heavy twin-turbine (HT) helicopter pilots. This was unexpected given that the pilots of MT helicopters rated risk higher for the VH phases (rejection of H08, Table 6). Therefore, it would have been natural that shorter time gaps would have been suggested by MT pilots. This contradiction might indicate that the risk experienced did not relate to the VH phases, that MT pilots had low awareness of practicing needs, or potentially displayed some ‘locus of control’ attitude(13,14,87,88) for example. It is also possible that such helicopters instigate complacent behaviour, which might in turn result in unjustifiably over-stretched envelopes(22). Given the potential for a detrimental impact on operational safety, the reasons for this discrepancy between risk rating and time gap for recency decay require further investigation.

H024 was rejected as glass cockpit helicopter pilots mentioned shorter time gaps for recency decay. Together with the rejection of H09 discussed in the previous section, this indicates that flying glass cockpits is not a safety panacea and might be more cognitively demanding than assumed in current training requirements(80). This is currently under investigation in Europe(32,38).

The rejection of H028 occurred as senior pilots (as opposed to flight line-only pilots) mentioned larger time gaps for recency decay (i.e., senior pilots’ mean = 52.1 days, SD = 75.9; flight line-only pilots’ mean = 6.2 days, SD = 7.2). This could stem from a sense that pilots with a larger overall experience can afford practicing night flights less often and still feel current, or might reflect a professional bias of senior pilots. By being involved with regulatory authorities and the OGP companies, senior pilots might have displayed a tendency to comply with the 90-day night-time recency standard currently applied in most countries(23,24,72).

The rejection of H035 resulted from the positive correlation between the number of night deck landings and the time gap for recency decay. The possible reasons for the rejection of H028 mentioned above are equally plausible. Because this significant correlation of H035 could be related to the rejection of H028 (i.e., pilots in senior positions would naturally have the largest numbers of night deck landings) and the potential for other confounding associations, a number of complementary meta-demographic tests were performed.

4.3.6 Meta-demographic analysis

Given the data characteristics (i.e., non-normality, heterogeneity of variance) and sample sizes involved, non-parametric independent samples Mann-Whitney U tests were used for the analyses across continuous and binomial categorical variables. The cross-tabulations between categorical variables were undertaken using the Pearson’s chi-square or Fisher’s exact tests.

The Pearson’s chi-square test investigates the null hypothesis that the categorical variables are independent when analysed in pairs. Based on the binomial distributions of marginal frequencies, this test compares the actual counts across the events of the variables with the counts that would be expected if the variables were independent. The greater the discrepancy, the larger is the dependency (also called ‘association’). The discrepancy is calculated by the chi-square statistic, as follows in Equation (7):

. . . (7)

where nij and μij are the observed and expected frequencies, and i and j represent rows and columns

18

The Pearson’s Chi-Square Test investigates the null hypothesis that the categorical variables are independent when analysed in pairs. Based on the binomial distributions of marginal frequencies, this test compares the actual counts across the events of the variables with the counts that would be expected if the variables were independent. The greater the discrepancy, the larger is the dependency (also called ‘association’). The discrepancy is calculated by the chi-square statistic, as follows in equation 7:

μμ

(7)

where nij and μij are the observed and expected frequencies, and i and j represent rows and columns in the contingency tables, respectively. This test requires that not less than 80% of the cells in such tables have expected frequencies greater than five and that no single expected frequency is lower than unity. When this was violated, Fisher’s Exact Test was used. This calculates exact probabilities based on the hyper geometric distribution of inner cell counts [11, 22, 81, 84]. Table 8 summarises the hypothesis tested and results obtained.

Table 8 – Statistical hypotheses, tests and results amongst demographic variables

The rejection of H036 confirmed the expectation that the senior pilots also had the largest numbers of nighttime deck landings. Coupled with the rejection of H028 and H035 mentioned above, it is possible that experience, as measured solely by the number of nighttime deck landings, and not by any form of flying hours, can be used to establish pilot nighttime recency requirement. This is partially recognised in the recency requirement of at least of 3 nighttime deck landings in a period of 90 days [72]. However, there is currently no adjustment of the requirement according to pilot operational experience.

H038 was rejected as the pilots of glass cockpit helicopters had the largest numbers of nighttime deck landings. This might reflect the gradual phasing-out of older helicopter types (thereby progressively performing fewer landings) or that a considerably high number of experienced pilots (i.e., with a large numbers of nighttime deck landings logged throughout their careers) now fly newer helicopter types. Because the data collected did not discriminate number of landings performed by pilots per aircraft type, the rejection of this hypothesis has little operational meaning.

The rejection of H040 occurred as a greater number of senior pilots (as opposed to flight line-only pilots) flew old helicopter types. This supports the rejection of H024, as senior pilots also tended to opt for larger time gaps for recency decay.

4.3.7 Preferred visual approach technique per demographic variables

Similar to Table 7, fourteen statistical hypotheses were tested (i.e., H042-H055). Apart from the Chi-Square, Fisher’s Exact and Independent Samples Mann-Whitney U Tests mentioned above, one Independent Samples t Test was used to test the significance of the differences in pilot age per preferred visual approach procedure. This was needed as the per-category distributions of age were normal and the variance was homogenous.

Formatted: Font: Italic


in the contingency tables, respectively. This test requires that not less than 80% of the cells in such tables have expected frequencies greater than five and that no single expected frequency is lower than unity. When this was violated, Fisher’s exact test was used. This calculates exact probabilities based on the hyper geometric distribution of inner cell counts(11,22,81,84). Table 8 summarises the hypothesis tested and results obtained.

The rejection of H036 confirmed the expectation that the senior pilots also had the largest numbers of night-time deck landings. Coupled with the rejection of H028 and H035 mentioned above, it is possible that experience, as measured solely by the number of night-time deck landings, and not by any form of flying hours, can be used to establish pilot night-time recency requirement. This is partially recognised in the recency requirement of at least of three night-time deck landings in a period of 90 days(72). However, there is currently no adjustment of the requirement according to pilot operational experience.

H038 was rejected as the pilots of glass cockpit helicopters had the largest numbers of night-time deck landings. This might reflect the gradual phasing-out of older helicopter types (thereby progres-sively performing fewer landings) or that a considerably high number of experienced pilots (i.e., with a large numbers of night-time deck landings logged throughout their careers) now fly newer helicopter types. Because the data collected did not discriminate number of landings performed by pilots per aircraft type, the rejection of this hypothesis has little operational meaning.

The rejection of H040 occurred as a greater number of senior pilots (as opposed to flight line-only pilots) flew old helicopter types. This supports the rejection of H024, as senior pilots also tended to opt for larger time gaps for recency decay.

4.3.7 Preferred visual approach technique per demographic variables

Similar to Table 7, 14 statistical hypotheses were tested (i.e., H042-H055). Apart from the chi-square, Fisher’s exact and independent samples Mann-Whitney U tests mentioned above, one independent

Table 8Statistical hypotheses, tests and results amongst demographic variables


samples t test was used to test the significance of the differences in pilot age per preferred visual approach procedure. This was needed as the per-category distributions of age were normal and the variance was homogenous.

The independent samples t statistic is calculated as in Equation (8):

. . . (8)

where X̄i is the mean value of group i, sp2 is the pooled variance that accounts for differences in

per-group sample sizes, and nj is the sample size of group j(81).The results of the statistical tests performed were all insignificant. This indicates that the choice

of specific visual approach technique tends to follow the pilot’s discretion only.

5.0 IMPLICATIONS AND CONCLUSIONSEven though ICAO recommends safety survey and task analysis as fundamental processes of hazard identification within the safety management framework, no guidance is provided on how to conduct surveys and task analyses. This paper offers, for the first time to the authors’ knowledge, a structured survey methodology for identifying the critical phases of offshore helicopter flights. The survey additionally outlines simple data collection methods that overcome the limitations of previous analyses, i.e., opinion- rather than experience-based, low representativeness, data incompleteness and use of clustered phases of flight, and thus generates statistically useful safety data. This is of utmost importance given the current scenario of few accidents (11,22), failed safety interventions based on accident data (26,56) and reported incidents that cannot be trusted as precursors to accidents(13,21).

Based on a novel and bespoke taxonomy which describes the phases of night-time helicopter flights beyond any existing taxonomy, careful questionnaire design and piloting, collection of data from key subject matter experts and the study of the data’s properties, multiple statistical hypotheses were tested. Confirming the well-documented operational experience presented throughout the paper, the results independently show that the phases flown by reference to external visual cues at high kinetic energy states (i.e., VH) cause the greatest concerns to pilots. Among such phases, the visual segment of instrument approaches is particularly problematic.

The analysis of risk levels experienced during this phase against the demographic variables failed to show any statistical significance. This is important as the sampling strategies of follow-up studies should be based on criteria other than the variables explored in this study. However, for the remaining VH phases (i.e., visual takeoff and go around), studies covering the operations of medium twin-turbine and glass cockpit helicopters should be given priority.

The perception of risk by medium twin-turbine (MT) helicopter pilots should be investi-gated further as such helicopters were previously associated with pilot-related and night-time accidents(11,22), higher risk ratings on the VH phases, but not with a perceived need for more frequent night-time flying practice. Likewise, recency requirements for glass cockpits require further investigation as they appear to differ from those currently applied, which were established for pilots of legacy cockpits with analogic and mixed designs.

The rejection of a number of interrelated hypotheses revealed that night-time recency require-ments might be better understood in light of a pilot’s number of night-time deck landings, as opposed to the flying hours logging regime typical of aviation. This is also corroborated by the

19

The Independent Samples t statistic is calculated as in equation 8:

� � ��

��

(8)

where �� is the mean value of group i, ��is the pooled variance that accounts for differences in per-group sample sizes, and nj is the sample size of group j [81].

The results of the statistical tests performed were all insignificant. This indicates that the choice of specific visual approach technique tends to follow the pilot’s discretion only.

5. Implications and conclusions

Even though ICAO recommends safety survey and task analysis as fundamental processes of hazard identification within the safety management framework, no guidance is provided on how to conduct surveys and task analyses. This paper offers, for the first time to the authors’ knowledge, a structured survey methodology for identifying the critical phases of offshore helicopter flights. The survey additionally outlines simple data collection methods that overcome the limitations of previous analyses, i.e., opinion- rather than experience-based, low representativeness, data incompleteness and use of clustered phases of flight, and thus generates statistically useful safety data. This is of utmost importance given the current scenario of few accidents [11, 22], failed safety interventions based on accident data [26, 56] and reported incidents that cannot be trusted as precursors to accidents [13, 21].

Based on a novel and bespoke taxonomy which describes the phases of nighttime helicopter flights beyond any existing taxonomy, careful questionnaire design and piloting, collection of data from key subject matter experts and the study of the data’s properties, multiple statistical hypotheses were tested. Confirming the well-documented operational experience presented throughout the paper, the results independently show that the phases flown by reference to external visual cues at high kinetic energy states (i.e., VH) cause the greatest concerns to pilots. Among such phases, the visual segment of instrument approaches is particularly problematic.

The analysis of risk levels experienced during this phase against the demographic variables failed to show any statistical significance. This is important as the sampling strategies of follow-up studies should be based on criteria other than the variables explored in this study. However, for the remaining VH phases (i.e., visual takeoff and go around), studies covering the operations of medium twin-turbine and glass cockpit helicopters should be given priority.

The perception of risk by medium twin-turbine (MT) helicopter pilots should be investigated further as such helicopters were previously associated with pilot-related and nighttime accidents [11, 22], higher risk ratings on the VH phases, but not with a perceived need for more frequent nighttime flying practice. Likewise, recency requirements for glass cockpits require further investigation as they appear to differ from those currently applied, which were established for pilots of legacy cockpits with analogic and mixed designs.

The rejection of a number of interrelated hypotheses revealed that nighttime recency requirements might be better understood in light of a pilot’s number of nighttime deck landings, as opposed to the flying hours logging regime typical of aviation. This is also corroborated by the meaningfulness of


meaningfulness of calculating helicopter pilot-related accident rates by cycles of takeoff and landing(11,22). Nevertheless, the 90-day recency standard generally practiced in the industry is considerably longer than the average 28 day time gap suggested by the pilots for assured recency. A combination of the current standard of three cycles of take-off/landing and the 28 day time gap identified will be the best starting point for focused experimental research aimed to determine appropriate recency requirements. Experimental research should potentially involve the observation of pilots in high-fidelity flight simulators, as well as the analysis of real FDM data.

As safety interventions are put in place to address phase-specific safety risks, it will be important to detect phases accurately and reassess the changes in associated risk levels. Precise phase discrimination might be attempted through enhancing FDM algorithms and by integrating FDM data with data from other sources, e.g., from cockpit voice and video recorders, for joint analysis. The statistical methods described in this paper will equally apply in the analysis of the risk levels attributed by pilots to more precisely discriminated phases of flight.

6.0 FUTURE ENHANCEMENTS OF THE SAFETY SURVEY METHODOLOGY

Surveys are necessarily based on subjective data such as content-dependent personal interpre-tations, experience, perceptions and information recollected from memory. Therefore, a degree of bias can never be discarded(13,23). Additionally, despite the attempted rigour of the sampling strategies, it could be argued that there was no randomisation of the study sample. This stems from the fact that the pilots were volunteers and hence self-selection and non-responder biases might have occurred(54). While these are fair concerns, the simplicity of the questions, the impartiality of the invitation letter and the focus on senior pilots with declarative knowledge and considerable professional experience are likely to have minimised any tendentious responses. These attributes should guide future survey attempts.

Representativeness can also be questioned, since a sampling frame was unavailable at the onset. However, conversations with the collaborating individuals in each company revealed that the percentage representativeness was high in at least the North Sea countries, Brazil and Spain (i.e., over 30% of the population of ‘senior pilots’ in each region). Hence, the results can be expected to be a fair representation of the corresponding regions and companies covered, which are in any case the main companies in the sector.

Finally, suboptimal sample sizes limited the analysis that could be undertaken, e.g., analysis by specific regions of operations was not possible. However, given the ethical imperative to keep survey participation voluntary, sub-optimal sample sizes often cannot be avoided. Therefore, the results should be interpreted with respect to such real-life constraints and dealt with as the first step towards evidence-based recommendations, rather than a fully validated study. Further experimental studies involving test flights are recommended.

ACKNOWLEDGEMENTSThe Lloyd’s Register Foundation and Santander Universities for funding, all the collaborating companies and pilots for the data, the SMEs for their invaluable criticism of the methods, and the reviewers for their knowledgeable and constructive critique.


REFERENCES1. khATwA, r. and roelen, A.l.c. An analysis of controlled-flight-into-terrain accidents of commercial

operations, 1988 through 1994, Flight Safety Digest, 1996, 17, (11-12), pp 166-212.2. helmreich, r.l. and khATwA, r. Analysis of critical factors during approach and landing in accidents

and incidents and normal flight, Flight Safety Digest, 1998, 18, (1-2), pp 1-77.3. hArris, f.D., kAsper, e.f. and iseler, l.e. US civil rotorcraft accidents, 1963 through 1997, 2000,

NASA, Moffett Field, CA, USA.4. JSSI – Occurrence Data Analysis Working Group, Analysis capabilities specification document –

Mandatory occurrence data, 2006, European Commission Joint Research Centre, Amsterdam, The Netherlands.

5. TAber, m. and mccAbe, J. Helicopter ditching: Time of crash and survivability. SAFE J, 2006, 34, (1), pp 5-10.

6. US JHSAT, Calendar Year 2000 Report to the International Helicopter Safety Team, 2007, US Joint Helicopter Safety Analysis Team, USA.

7. JArVis, s. and hArris, D. Investigation into accident initiation events by flight phase for highly inexpe-rienced glider pilots, Int J Applied Aviation Studies, 2008, 8, (2), pp 211-224.

8. menzel, R. ECCAIRS SC 2011 – EASA update, 2011, EASA, Cologne, Germany.9. pAng, M. Aviation safety data collection and processing (SDCP) – ECCAIRS Software, 2011 [cited

2013 11/10]; http://www.asc.gov.tw/media_files/31_pap.pdf.10. bArroso, J.m. Commission Regulation (EU) No 965/2012 of 5 October 2012 – laying down technical

requirements and administrative procedures related to air operations pursuant to Regulation (EC) No 216/2008 of the European Parliament and of the Council, European Commission, Editor, 2012, Official J European Union, Brussels, Belgium.

11. nAscimenTo, f.A.c., mAJumDAr, A. and ochieng, w.y. Helicopter accident analysis. J Navigation, 2014. 67, (1), pp 145-161.

12. nAscimenTo, f.A.c., mAJumDAr, A. and ochieng, w.y. Investigating the truth of Heinrich’s pyramid in offshore helicopter transportation, Transportation Research Board 92nd Annual Meeting, 2013, TRB, Washington, DC, USA.

13. nAscimenTo, f.A.c., mAJumDAr, A. and ochieng, w.y. Investigating the truth of Heinrich’s pyramid in offshore helicopter transportation, Transportation Research Record: J Transportation Research Board, 2013, 2336, pp 105-116.

14. nAscimenTo, F.A.C. eT Al A multistage multinational triangulation approach to hazard identification in night-time offshore helicopter operations, Reliability Engineering & System Safety, 2012, 108, (0), pp 142-153.

15. hoh, r.h. eT Al Decision-height windows for decelerating approaches in helicopters – Pilot/vehicle factors and limitations, DOT/FAA/CT-90/14, 1991, FAA, VA, USA.

16. CAA, CAP 382 – The Mandatory Occurrence Reporting Scheme. 2005, CAA, Gatwick, UK.17. EASA, Annual Safety Review 2008, 2009, EASA, Cologne, Germany.18. mAJumDAr, A. eT Al A causal factors analysis of helicopter accidents in New Zealand 1996-2005 and

the United Kingdom 1986-2005, Aeronaut J, 2009, 113, (1148), pp 647-660.19. US JHSAT, The Compendium Report: The US JHSAT Baseline of Helicopter Accident Analysis, Volume

I (CY2000, CY2001, CY2006), 2011, US Joint Helicopter Safety Analysis Team.20. Joint Research Centre of the European Commission. ECCAIRS 4.2.8 Data Definition Standard – Event

phases, 2012, [cited 2013 10/04],http://eccairsportal.jrc.ec.europa.eu/index.php/Products/83/0/.21. nAscimenTo, f.A.c., mAJumDAr, A. and ochieng, w.y. Incident reporting in offshore helicopter transpor-

tation, 38th European Rotorcraft Forum, 2012, Amsterdam, The Netherlands.22. nAscimenTo, f.A.c., mAJumDAr, A. and ochieng, w.y. A 15-year multivariate analysis of worldwide

offshore helicopter accidents, AHS 69th Annual Forum and Technology Display, 2013, AHS, Phoenix, AZ, USA.

23. nAscimenTo, f.A.c., JArVis, s. and mAJumDAr, A. Factors affecting safety during night visual approach segments for offshore helicopters, Aeronaut J, 2012, 116, (1175), pp 303-322.

24. nAscimenTo, f.A.c., mAJumDAr, A. and JArVis, s. Nighttime approaches to offshore installations in Brazil: Safety shortcomings experienced by helicopter pilots, Accident Analysis & Prevention, 2012. 47, (0), pp 64-74.

25. CAA, Helicopter flight in degraded visual conditions, CAA Paper 2007/03, 2007, CAA, London, UK.


26. sTeVens, m. Aviation safety in the oil & gas business, 2013, CHC Safety and Quality Summit, Vancouver, Canada.

27. olsen, O.E. and linDøe, P.H. Risk on the ramble: The international transfer of risk and vulnerability, Safety Science, 2009, 47, (6), pp 743-755.

28. Vinnem, J.E. Risk indicators for major hazards on offshore installations, Safety Science, 2010, 48, (6), pp 770-787.

29. Vinnem, J.E. Evaluation of offshore emergency preparedness in view of rare accidents, Safety Science, 2011. 49, (2), pp 178-191.

30. OGP, Safety Performance Indicators – 2011 data, 2012, International Association of Oil & Gas Producers.31. oksTAD, e., Jersin, e. and TinmAnnsVik, r.k. Accident investigation in the Norwegian petroleum

industry – Common features and future challenges, Safety Science, 2012, 50, (6), p 1408-1414.32. CAA, CAP 1145 – Safety review of offshore public transport helicopter operations in support of the

exploitation of oil and gas, 2014, UK Civil Aviation Authority, London, UK.33. howson, D.A. Enhancing offshore helideck lighting, Maritime Operations of Rotorcraft, 2008, RAeS:

London, UK.34. CAA, Hazard analysis of the use of GPS in offshore helicopter operations, 2010, CAA, Gatwick, UK.35. CAA, CAP 437 – Offshore helicopter landing areas – Guidance on standards, 2010, CAA, Gatwick, UK.36. CAA, CAA Paper 2010/01 – The SBAS offshore approach procedure (SOAP), 2010, CAA, Gatwick, UK.37. howson, D.A. Research initiatives for improving the safety of offshore helicopter operations, Aeronaut J,

2006, 110, (1113), p 14.38. EASA, Notice of Proposed Amendment 2013-10 – Helicopter Offshore Operations – RMT.0409 &

RMT.0410 (OPS.093(a)&(b)) – 06/06/2013, 2013, EASA, Cologne, Germany.39. keATing, C. Stabilised approaches at night, 2009, CAA, Gatwick, UK.40. howson, D. Review and analysis of UK and European Part 27 Helicopter incident and accident data,

Sixth EASA Rotorcraft Symposium, 2012, EASA, Cologne, Germany.41. CAST/ICAO. Phase of Flight – Definition and Usage Notes. 2011, [cited 2012 10/06], http://www.

intlaviationstandards.org/Documents/PhaseofFlightDefinitions.pdf.42. TeixeirA, S., Helicopter flight operational quality assurance (HFOQA): Development of HFOQA analysis

software, 2006, Embry-Riddle Aeronautical University, Daytona Beach, FL, USA.43. kirwAn, b. and AinsworTh, l.k. A Guide to Task Analysis, 1992, Taylor and Francis, London, UK.44. sTolzer, A.J., hAlforD, c.D. and gogliA, J.J. Safety Management Systems in Aviation, Ashgate Studies

in Human Factors for Flight Operations, 2008, Ashgate, Aldershot, UK.45. Skybrary. Flight phase taxonomy, 2010, [cited 2012 05/06], http://www.skybrary.aero/index.php/

Flight_Phase_Taxonomy.46. ross, c. and gibb, g. A risk management approach to helicopter night offshore operations, 2008

[cited 2010 15/10], http://asasi.org/papers/2008/Risk%20Approach%20to%20Night%20Offshore%20Operations%20Presented%20by%20Gerry%20Gibb%20&%20Cameron%20Ross.pdf.

47. couch, m. and linDell, D. Study on rotorcraft safety and survivability, 2010, AHS 66th Annual Forum, AHS, Phoenix, AZ, USA.

48. herrerA, I.A. eT Al Helicopter safety study 3, Main Report, 2010, SINTEF, Trondheim, Norway.49. morrison, G., Helicopter Safety Offshore, 2000/089, 2001, Health and Safety Executive, Aberdeen, UK.50. brooks, C.J. eT Al Civilian helicopter accidents into water: Analysis of 46 cases, 1979-2006, Aviation,

Space and Environmental Medicine, 2008, 79, (10), pp 935-940.51. ICAO, The Convention on International Civil Aviation – Annex 19 – Safety Management, 2013, ICAO,

Montréal, Canada.52. ICAO, Safety Management Manual (SMM), 2009, ICAO, Montréal, Canada.53. boscherT, g., Dery, m. and husTwiT, J. Pilot safety survey 2005: What pilots are saying, Air Medical J,

2007, 26, (1), pp 34-37.54. Dery, m., eT Al Results and recommendations from the helicopter EMS pilot safety survey 2005, Air

Medical J, 2007, 26, (1), pp 38-44.55. konTogiAnnis, T. and mAlAkis, s. A systemic analysis of patterns of organizational breakdowns in

accidents: A case from Helicopter Emergency Medical Service (HEMS) operations, Reliability Engineering & System Safety, 2011, 99, (0), pp 193-208.

56. huber, m. Airline safety system comes to helicopter sector, 2013, [cited 2013 3/5], http://www.ainonline.com/aviation-news/aviation-international-news/2013-05-03/airline-safety-system-comes-helicopter-sector.

57. ATSB, Perceived pilot workload and perceived safety of RNAV (GNSS) approaches, 2006, ATSB, Canberra, Australia.


58. suboTic, b., ochieng, w.y. and sTrAeTer, o. Recovery from equipment failures in ATC: Determination of contextual factors, Reliability Engineering & System Safety, 2007, 92, (7), pp 858-870.

59. CAA, CAA Paper 97009 – A questionnaire survey of workload and safety hazards associated with North Sea and Irish Sea helicopter operations, 1997, CAA, London, UK.

60. miTchell, s.J. and brAiThwAiTe, g.r. Perceptions of safety and offshore helicopter travel, Int J Energy Sector Management, 2008, 2, (4), pp 479-498.

61. Canada-Newfoundland and Labrador Offshore Petroleum Board, Offshore Helicopter Safety Enquiry, 2010, p 324, St John’s, Newfoundland and Labrador, Canada.

62. bArneTT, r.D. Middle East-(GCC)-JHSAT Safety Team Overview Briefing. 2009, [cited 2010 20/11]; Available from: http://www.ihst.org/portals/54/montreal/GULFOPERATION.pdf.

63. sriDhArAm, K. Update on Activities of JHSAT (India), 2009, [cited 2010 Web Page], http://www.ihst.org/portals/54/montreal/AVMKSRIDHARAN.pdf.

64. burns, r.p. and burns, r.A. Business Research Methods and Statistics Using SPSS, 2008, SAGE Publications, London, UK.

65. simons, r., wilschuT, e.s. and VAlk, p.J.l. Sleep and alertness in North Sea helicopter operations, Aviation, Space and Environmental Medicine, 2011, 82, (7).

66. ebbATson, M. The loss of manual flying skills in pilots of highly automated airliners, Systems Engineering and Human Factors, 2009, Cranfield University, UK.

67. nAscimenTo, f.A.c., mAJumDAr, A. AnD ochieng, w.y. Nighttime offshore helicopter operations – Identi-fication of contextual factors relevant to pilot performance, Advances in Human Aspects of Aviation, 2013, pp 224-236, lAnDry, S.J., Editor, CRC Press, Boca Raton, USA.

68. smiTh, p.c. and kenDAll, l.M. Retranslation of expectations: an approach to the construction of unambiguous anchors for rating scales, J Applied Psychology, 1963, 47, (2), pp 149-155.

69. rollenhAgen, C. eT Al The context and habits of accident investigation practices: A study of 108 Swedish investigators, Safety Science, 2010, 48, (7), pp 859-867.

70. ICAO, Safety Management Manual (SMM), 2012, ICAO, Montréal, Canada.71. likerT, R. A Technique for the measurement of attitudes, Archives of Psychology, 1932, 140, pp 1-55.72. OGP, Aircraft Management Guidelines, 2013, International Association of Oil & Gas Producers, London, UK.73. nAscimenTo, f.A.c., mAJumDAr, A. and ochieng, w.y. Nighttime offshore helicopter operations –

identification of contextual factors relevant to pilot performance, 2012, 4th International Conference on Applied Human Factors and Ergonomics (AHFE), San Francisco, USA.

74. IHST. The developing history of an accident-free future, 2011, [cited 2013 04/11], http://www.ihst.org/Default.aspx?tabid=1780&language=en-US.

75. corbeT, A.T. and AnDerson, J.r. Knowledge Tracing: Modelling the Acquisition of Procedural Knowledge. User Modelling and User-Adapted Interaction, 1995, 4, pp 253-278.

76. chipmAn, s.f., schrAAgen, J.m. and shAlin, V.l. Introduction to cognitive task analysis, Cognitive task analysis, 2000, pp 3-23, schrAAgen, J.m. chipmAn, s.f. and shAlin, V.l., Editors, Lawrence Erlbaum Associates, Mahwah, NJ, USA.

77. JAA, JAR-OPS 3: Commercial Air Transportation (Helicopters), 2007, JAA, The Netherlands.78. ANAC, Regulamento Brasileiro de Aviação Civil RBAC, 2010, 135, (00), ANAC, Brasilia.79. frAkes, m.A., high, k. and sTocking, J. Transport nurse safety practices, perceptions and experiences:

The Air and Surface Transport Nurses Association Survey, Air Medical J, 2009, 28, (5), pp 250-255.80. hArris, D., Head-down flight deck display design, Human Factors for Civil Flight Deck Design, 2004,

hArris, D., Editor, Ashgate, Aldershot, UK.81. fielD, A., Discovering Statistics Using SPSS, Third edition, 2009, SAGE, London, UK.82. IBM Corporation. Friedman’s Test (nonparametric tests algorithms), 2011, [cited 2013 08/05/], http://

publib.boulder.ibm.com/infocenter/spssstat/v20r0m0/index.jsp?topic=%2Fcom.ibm.spss.statistics.help%2Ftwo_independent_samples_test_types.htm.

83. IBM Corporation, Wilcoxon matched-pairs signed-rank test (nonparametric tests algorithms), 2011, [cited 2013 08/05/], http://publib.boulder.ibm.com/infocenter/spssstat/v20r0m0/index.jsp?topic=%2Fcom.ibm.spss.statistics.help%2Ftwo_independent_samples_test_types.htm.

84. AgresTi, A. Categorical Data Analysis, Second edition, 2002, Wiley Series in Probability and Statistics, Wiley, Hoboken, USA.

85. IBM Corporation. Mann-Whitney test (nonparametric tests algorithms), 2011, [cited 2013 06/03]; Available from: http://publib.boulder.ibm.com/infocenter/spssstat/v20r0m0/index.jsp?topic=%2Fcom.ibm.spss.statistics.help%2Falg_nonparametric_independent_mann-whitney.htm.

86. SAS Institute, Base SAS® 9.2 Procedures Guide: Statistical Procedures, 2010, Cary, NC, USA.


87. VAn Der schAAf, T. and kAnse, l. Biases in incident reporting databases: an empirical study in the chemical process industry, Safety Science, 2004, 42, (1), pp 57-67.

88. VAn Der schAAf, T. and kAnse, l. Checking for biases in incident reporting, Accident Precursor Analysis and Management: Reducing Technological Risk Through Diligence, 2004, phimisTer, J.r., bier, V.m. and kunreuTher, h.c., Editors, National Academy of Engineering, Washington, DC, USA.