Comparing gliding accident causes between inexperienced and experienced glider pilots

Differences in accident rates between under-10-hour and over-10-hour glider pilots

This research was carried out in the United Kingdom by Jarvis and Harris, using accident data ranging from 2002 to 20061.

In their work, they created a modified human factors model specifically for describing gliding accidents, arguing that existing models applied to general aviation did not adequately categorise such data (e.g. HFACS – Shappell and Weigmann 1990[2]). They used their model to analyse gliding accidents, which is the focus of this review.

The basic findings were as follows, shown in Illustration 1:

Illustration 1: Primary accident causation categories, with examples of two most common accident types in each category
Primary causal category: Perceptual Judgement Handling Strategy Attention
Total number of accidents in each category (N=355): 92 93 86 84
Total number of injuries in each category: 9 26 19 13
Examples of most common accident causes in each category: 1) Landing flare too low/too late; 2) Landing flare too high/too early 1) Failed to maintain (or increase to) speed required; 2) Allowed/failed to prevent wing going down 1) Flew out of reach of airfield during intended local flight; 2) Landed in an unsuitable field 1) Secondary control action omitted (unintentionally); 2) Secondary control action slip

Illustration 1 shows these major accident categories, with two examples of the most frequent lower-level accident causes in each.

They used the measure of number of launches per accident as an indication of accident rates.

The authors' data show that in every primary accident causation category, except for strategy, low-hour pilots had fewer number of launches per accident than did high-hour pilots. In other words, low-hour pilots had significantly higher numbers of accidents per number of launches. Specifically, the data show that there were significantly more perceptual judgement accidents among low-hour pilots (z = 2.9; p < 0.01), as well as significantly fewer concerning strategy (z = -2.6; p < 0.01). Presumably the second statistic represents the fact that under-10-hour pilots are not usually doing cross-country soaring.

Overall, under-10-hour pilots had at least 8.9 times the perceptual judgement accident rates of over-10-hour pilots, as well as at least 4.9 times the handling and 2.3 times the attention accident rates. The authors calculated these ratios between the upper bound of the under-10-hour 95% CI, and the over-10-hour base value.

Among the accident causes classified within each of the four top-level variables, some significant differences between low-hour pilots and higher-hour pilots were also found. Examples of two of these were landing flare too high or early, and insufficient airbrake reduction. Both of these statistics showed that low-hour pilots had much higher odds of performing these errors (OR = 6.73; p = 0.01, and OR = 19.74; p = 0.01 respectively[4]). Differences between low-hour and higher-hour pilots on two other of the most commonly-occurring variables (landing flare too late or low and too little airspeed) were not significantly different.[5]


Methods

Research approach

The researchers used quantitative analysis to attempt to demonstrate the differences in accident rates and types between low-hour (10 hours or less) and higher-hour glider pilots.

Sample

469 gliding accident reports were used as the raw data (these were sourced from the British Gliding Association’s database). These were narrowed down after coding to 359 accidents, less four whose cause could not be determined or were caused by a passenger. Thus 355 accidents were deemed pilot-related.

Design

This was exploratory, descriptive research, using what the authors refer to as a “data-driven approach”, making this a basically inductive design.

Variables

Four top-level variables were studied, representing primary categories of accident causation. These were: Perceptual judgement, handling, strategy, and attention.

Procedure

The authors worked with these four variables, which they had previously created using open coding and axial coding (Strauss and Corbin 1990[3]). These four variables were used to classify and compare accident rates between low-hour and higher-hour pilots.

Data analysis

The authors used a chi-squared test (Fisher’s exact test) to show any significant differences in the different accident rates between those participants with under 10 hours’ experience, and those over 10 hours, across the four top-level variables described above.

Generalization potential

The findings in this article are most directly applicable to UK glider pilots, and probably glider pilots in general.

Given that there is relatively little human factors research focused directly on glider pilots, the findings are potentially useful for training curricula, etc. However, because the findings were based on a newly-customised model, there is good motivation for the results to be replicated in order to enhance their validity.

References
1. JARVIS Steve & Don HARRIS (2006). Development of a bespoke human factors taxonomy for gliding accident analysis and its revelations about highly inexperienced UK glider pilots. Ergonomics, 2010, volume 53, number 2, pages 294-303.
2. SHAPPELL Scott and WEIGMANN Douglas (2000). The Human Factors Analysis and Classification System–HFACS: Final report. DOT/FAA/AM-00/7. Washington, DC: Office of Aviation Medicine.
3. STRAUSS Anselm and CORBIN Juliet (1990). Basics of qualitative research: Grounded theory procedures and techniques. Thousand Oaks, CA: Sage Publications.
+++ Notes +++
4. Even taking into account the large size of the 95% confidence interval in the frequency of airbrake misuse, the authors indicate that low-hour pilots were at the very least up to three times more likely to commit this error than higher-hour pilots (Jarvis and Harris 2010: p.299[1]).
5. Because of the small numbers and the accompanying loss of statistical power involved in some comparisons, there is a possibility that this is a Type 2 error, despite use of Fisher's exact test (which is designed for low expected cell counts). There is also risk of Type 1 error with the significant findings, for the same reason. Further research may shed light on whether this is so.

Contributors to this page

Authors / Editors

DAtkinsonDAtkinson


Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License