U. S. Department of Transportation
Download 2.66 Mb.
|
- Navigate this page:
- Day/Night Indicator
- Collision
Other VariablesThese variables do not necessarily fall into the other categories above. Snow Removal Vehicle Involved(Runway Incursion Database) This variable indicates whether a snow removal vehicle was involved in the event. This variable incorporates many effects under one umbrella: decreased visibility to snow, special operating procedures to accommodate snow removal and weather, and unfamiliar drivers with access to runways. It is not possible to disentangle these without more accurate measures of the component factors, such as driver experience or (especially) weather / visibility. Figure 42 presents the overall distribution of this variable. 
Figure – Distribution of Snow Removal Vehicle Involved Table 168 and Table 169 present the distribution of this variable by severity, while Table 170 and Table 171 present the distribution by incident type. Table – Observed Distribution of Snow Removal Vehicle Involved by Severity
Table – Expected Distribution of Snow Removal Vehicle Involved by Severity
Table – Observed Distribution of Snow Removal Vehicle Involved by Incident Type
Table – Expected Distribution of Snow Removal Vehicle Involved by Incident Type
The distribution by severity, and its associated Fisher’s Exact test statistic, indicates no relationship between severity and snow removal vehicles. While there are a relatively low number of observations, no dramatic trend by severity presents itself. This could be due to the fact that current operational changes when snow removal vehicles are present already compensate for the increased risk introduced. The distribution by incident type is more interesting. Firstly, the Chi-Squared statistic indicates that there is a relationship between the presence of snow removal vehicles and type. There are approximately 3 times as many observed V/PD incidents than expected. PD incidents are dramatically under-represented while OE incidents are close to their expected value. The large number of V/PD incidents is interesting, indicating that when snow removal vehicles are involved in an incident, they are disproportionally at fault. Given the high concentration of V/PD incidents, it is instructive to examine the severity of those incidents more closely. Recall that Table 168 indicated no relationship between severity and the presence of snow removal vehicles. That test statistic was calculated for all incident types, whereas Table 172 and Table 173 present the same information, distribution by severity, but only for V/PD incidents. Table – Observed Distribution of Snow Removal Vehicle Involved by Severity, V/PD Only
Table – Expected Distribution of Snow Removal Vehicle Involved by Severity, V/PD Only
Future Research
Here, the test statistic indicates that there is a relationship among severity. Category D appears to be underrepresented while the conflict categories are all overrepresented. This indicates that V/PD incidents involving snow removal vehicles tend to be more severe than V/PD incidents not involving snow removal vehicles. The trend among conflict incidents is less clear (partly due to sample size issues). It is possible that snow removal vehicles are more likely to conflict with aircraft than other types of vehicles due to the nature of their operations. A better examination of the involvement in snow removal vehicles would account for the fact that snow removal vehicles are some of the few vehicles operating on runways. Further investigation is necessary to determine if snow removal vehicles are actually more risky or their over representation in conflict events is a product of their unique activities. Day/Night Indicator(Runway Incursion Database) This variable indicates if the event occurred during the daytime. As the hours of daylight shift throughout the year, this is perhaps a better (though slightly subjective) measure than the hour the incident occurred. This variable originates from the Runway Incursion database and is thus available for a large number of incidents. Figure 43 presents the overall frequency of this day/night indicator (note that a coding of “yes” indicates daytime). 
Figure – Overall Frequency of Day/Night Indicator Table 174 and Table 175 present the distribution of this variable by incident severity. Table – Observed Distribution of Day/Night by Severity
Table – Expected Distribution of Day/Night by Severity
As daytime and nighttime are opposites, it may be more instructive to examine the “No” row above; that is, observations coded as “No” for daytime must, by definition, have occurred after dark. The test statistic indicates that there is indeed a relationship between daytime/nighttime and severity. Examining the expected values indicates that categories A, B, and C are overrepresented at night while category D is underrepresented. This suggests that conflict incidents are more likely to occur at night. Table 176 and Table 177 present the distribution by incident type. Again, there is a significant relationship between these two variables. OE and V/PD incidents occur more often than expected at night, while PD incidents occur less frequently than expected at night. This may be due to macroscopic patterns in pilot behavior throughout the day. Less experienced pilots may not be (or be allowed to be) flying at night and thus are unable to commit errors. Given the strong relationship between incident type and severity, it is possible that the severity relationship seen in Table 176 is a product of the relationship of incident type. Further research into the relationship between day/night and severity should account for incident type explicitly. Additionally, it is unclear why day/night would impact the three incident types differently. An examination of into these differing impacts and how they may contribute to severity would help better understand the impact of day/night on runway incursions. Table – Observed Distribution of Day/Night by Incident Type
Table – Expected Distribution of Day/Night by Incident Type
Events occurring at night have odds of being severe of approximately 83% higher than those occurring in daytime. As indicated in Table 178, this result is fairly precise. A similar result holds for the impact on the likelihood of being an OE (compared to either PD or V/PD), as seen in Table 179. Interestingly, the effects are approximately the same size. It is possible this similarity is driven by the underlying relationship between incident type and severity. The results presented in Table 180 attempts to correct for this. Table – Logit Estimate of Impact on Severity, Night
Table – Logit Estimate of Impact on Incident Type, Night
Table – Logit Estimate of Impact on Night55
Interestingly, the effects persist. That is, night impacts severity, even when accounting for incident type, and night also impacts incident type even when accounting for severity. These results also tell us two further things. Firstly, the size of the impacts is indistinguishable (the difference of the coefficients is not significantly different from zero). Secondly, there is no interaction effect. That is, night makes an incident more likely to be severe and more likely to be an OE, but only as the sum of its parts. Another way to think about it is that the odds ratios are multiplicative: night increases the odds of a severe OE by approximately 2.9 (1.7 * 1.7 = 2.9). That the effects are relatively constant in size over multiple model specifications and are precisely estimated indicates that this is likely a robust impact. Further research into the exact mechanism through which night impacts severity and controller actions may yield results that could improve operations. Future Research
Collision(Runway Incursion Database) Collisions between aircraft are also tracked in the Runway Incursions database, provided they occur on a runway. While exceedingly rare (only 7 appear in the 10 years covered by the dataset), it may be helpful to examine these incidents. Note that all collisions are considered a category A incursion, so no analysis of severity is possible. Table – Logit Estimate of Impact on Likelihood of Collision, OE Incident
Table 181 indicates the increase in the odds of a collision, given that the event is an OE (the alternative being PD or V/PD). While the increase is quite dramatic (almost 15 times as high as non-OE incidents), the confidence interval is also quite large. It is important to consider the variance in the estimate as well as the magnitude of the estimate. There is little doubt that an OE incident has higher odds of being a collision, but the odds may increase anyway from approximately 2 to 77 times. Due to the extreme rarity of collision events, it will be difficult to get a more precise estimate without much more data, which is, in this case, not a desired event. This result further supports the claim that OE incidents tend to be more severe, but more research into why OE incidents are more severe is still required.
|