U. S. Department of Transportation
Download 2.66 Mb.
|
- Navigate this page:
- Foreign Aircraft or Pilot
- Pilot Lost
- Pilot Ratings
- Entered Runway without Clearance
- Pilot Instrument Rating
- Pilot Hours in Make and Model
Pilot InformationThis information describes the pilot involved in the incident. This information comes from the Runway Incursion and ATQA PD databases. Some variables may only pertain to PD incidents, which are noted in the variable specific discussion. Some variables are categorical while others are continuous. Foreign Aircraft or Pilot(Runway Incursion Database) This variable indicates whether or not the pilot or aircraft involved were of foreign origin. Table 49 and Table 50 provide the breakdown of this variable by severity. Figure 8 presents the overall distribution of this variable. 
Figure – Distribution of Foreign Pilot Status Foreign pilots are weighted toward conflict categories… [because most] are commercial pilots The distribution is weighted towards the conflict categories of A, B and C. The test statistic indicates that there is a relationship between these two variables. An underlying cause may be that most foreign pilots or aircraft entering the United States are commercial. Commercial pilots are (generally) at busier airports, and so are less likely to be in a category D due to the increased traffic at the airport. Because of the strong relationship between foreign pilot status and commercial carrier status, it is difficult to draw strong conclusions about the effect of foreign pilot status on severity. Table – Observed Distribution of Foreign Aircraft or Pilot by Severity
Table – Expected Distribution of Foreign Aircraft or Pilot by Severity
Pilot Lost(ATQA PD) This variable indicates whether the investigation determined if the pilot was lost at the time of the incident. Figure 9 presents the overall distribution of this variable. Table 51 and Table 52 present a tabulation of this variable by severity. Fisher’s Exact test indicates that there is no relationship between these variables. While not entirely surprising, this does indicate that, at least at a cursory level, pilots being lost on the airfield does not increase the severity of an ensuing incident. It may, however, increase the likelihood of an incident occurring at all; this cannot be tested without “normal operations” data for all non-incident operations. 
Figure – Distribution of Pilot Lost Table – Observed Distribution of Pilot Lost by Severity
Table – Expected Distribution of Pilot Lost by Severity
Pilot Ratings(ATQA PD) The ATQA PD database contains information on pilot ratings. These ratings include: single engine sea, single engine land, multiengine sea, multiengine land, rotorcraft, glider, lighter than air, and other. The sea and land ratings for multiengine and single engine categories were combined due to the low number of sea plane ratings in the dataset. The distribution of response after the combination of sea and land ratings can be seen in Figure 10. Chi-Squared tests were run for each category separately; the majority of the categories have no significant relationship with severity. The only category that presented a marginally significant result was the multiengine rating category. The results of this test are presented in Table 53 and Table 54. The pattern of expected versus observed is unclear. The major contribution to the test statistic appears to be from the overrepresentation of category C incursions. This may be an artifact of the distribution of multiengine rating in the population; that is, pilots with multiengine ratings may fly into busier airports and thus would be more likely to be in a conflict event. 
Figure – Frequency of Pilot Ratings by Rating Category Table – Observed Distribution of Multiengine Rating by Severity
Table – Expected Distribution of Multiengine Rating by Severity
Entered Runway without Clearance(Runway Incursion Database) If the primary aircraft in the event entered a runway without clearance, this variable is coded yes. The Chi-Squared statistic, contained in Table 55 and Table 56, indicates that there is a relationship between this variable and severity. Category C is underrepresented while all other categories are over represented. 
Figure – Distribution of Entered Runway without Clearance Table – Observed Distribution of Entered Taxiway or Runway without Clearance by Severity
Table – Expected Distribution of Entered Taxiway or Runway without Clearance by Severity
Pilot Instrument Rating(ATQA PD) This variable indicates the instrument rating of the pilot involved in the database. Interestingly, the coding on this variable contains information on if the pilot was rated previously, but is not currently. Figure 12 presents the overall distribution of this variable. Table 57 and Table 58 present a breakdown of this variable. Note that unknown ratings were excluded, as they provide little insight into the impact of this variable. 
Figure – Distribution of Pilot Instrument Rating Table – Observed Distribution of Pilot Instrument Rating by Severity
Table – Expected Distribution of Pilot Instrument Rating by Severity
Table – Difference between Observed and Expected of Pilot Instrument Rating by Severity
The Chi-Squared test statistic indicates that there is a relationship between severity and instrument rating. Current and Not Current are underrepresented for categories A and B, see Table 59 for the deviations between observed and expected. The opposite is true for No Rating. For categories C and D, Not current and No Rating are underrepresented for category C and overrepresented for category D, while Current is observed more than expected for category C and less than expected for category D. When restricted to only conflict events (Table 60 and Table 61), Current and Not Current follow a similar pattern in terms of observed compared to expected values and the mitigating impact on severity is still present. However, the impact of having a non-current rating may be non-linear. These data suggest that having ever been rated is associated with lower incident severity. Without additional statistical and human factors study, it is unclear if these pilots get into fewer severe situations, better recover from mistakes, or if this can be explained by other factors, including the possibility of a spurious correlation. Specifically, this variable is easily conflated with commercial carrier status (as all commercial carriers are instrument rated while not all GA pilots are instrument rated). Table – Observed Distribution of Pilot Instrument Rating by Severity, Conflict Only
Table – Expected Distribution of Pilot Instrument Rating by Severity, Conflict Only
Finally, Table 62 presents an estimate of the impact on severity. The baseline here is considered to be a pilot with no instrument rating. Thus, the odds ratios represent the reduction in the likelihood of being severe if the pilot were either currently rated or rated in the past, but not currently. These estimates support the conclusion that pilots with either current rating or a past rating are less likely to be involved in a severe incursion. Interestingly, the confidence intervals for the two estimates overlap, indicating that the magnitude of the two estimates cannot be considered statistically different. That is, this preliminary estimate suggests that pilots with past ratings are as safe as pilots with current ratings. Further research into pilot instrument ratings should account for the three rating groups (current, past, and never rated) and further investigate whether current and past ratings have the same impact on severity. Table – Logit Estimate of Impact on Severity, Pilot Instrument Rating
Pilot Hours in Make and Model(ATQA PD) For each PD incident, pilots are required to report hours in the make and model of aircraft in which the incident occurred. Table 63 presents the percentiles of this distribution. While the overall distribution is interesting, the distribution of pilot hours by severity level is more pertinent to the question at hand. Figure 13 presents this distribution. Table – Percentiles of Pilot Hours in Make and Model
Figure – Distribution of Pilot Hours in a Make and Model [N]o pilot with more than 5,000 hours in the make and model… has committed a severe incursion Figure 13 presents two pieces of information. A histogram representing the distribution detailed in Table 63 is on the left. The graph on the right presents the distribution by severity in terms of a boxplot (or box and whisker plot).43 Figure 13 reveals that no pilot with more than 5,000 hours in the make and model involved in the incident has committed a severe incursion (category A or B). Figure 13 also reveals that the distribution of hours is weighted heavily towards zero for all severities. It appears, however, that the median value for categories A and B are lower than those of C and D, indicating a leftward shift in the distribution. In other words, pilots involved in category C and category D incursions tend to have more hours in that make and model. There are two possible explanations that come to mind. Appendix A:The most obvious is that there is an effect of experience. As pilots spend more hours in a make and model they are less likely to commit serious incursions. Appendix B:An alternative explanation is that bad pilots do not ever get many hours in a make and model. Under this hypothesis, error rates are fairly constant across experience levels but pilots that commit many serious errors stop being pilots (e.g., they do not enjoy it, cannot get licensed). This would lead to lower hour pilots being concentrated in categories A and B rather than in C or D. Future Research
Further statistical testing is required to distinguish between these two hypotheses. The two hypotheses also suggest different policy responses. One implies a policy to encourage pilots to get more hours more quickly. The other hypothesis implies that there needs to be a better way to identify poor pilots and remove them from the population. The medians of a continuous variable separated by groups can be compared using what is called the Kruskal-Wallis rank test.44 Table 64 reports the results of a Kruskal-Wallis test on pilot hours in the involved make and model. The test statistic indicates that the four severity categories are jointly different from each other. However, the pairwise comparisons indicate that no two groups can be considered different from each other (note that a stricter criterion has been used to determine significance given the multiple comparison issue noted in Appendix C.3). Table – Kruskal-Wallis Test Results for Pilot Hours in Make and Model
Future Research
In Figure 13 categories A and B appear similar as do categories C and D. Grouping the categories in this manner produce a dichotomized variable, which can be easier to analyze; some of the techniques in Section 4 rely on this dichotomous nature. On the other hand, grouping categories together causes a loss of information. In this case, it is no longer possible to distinguish between conflict and non-conflict events. Thus, while Table 65 presents the results of such a dichotomous grouping, further investigation into the differences between categories (especially categories C and D) is warranted. Table : Kruskal-Wallis Test Results for Pilot Hours in Make and Model, Severe versus Non-Severe
|