U. S. Department of Transportation
Download 2.66 Mb.
|
AirportThis set of models examines the physical characteristics of the airport at which the incursion occurred. It is important to note that the variables in these models do not vary by incursion (in general). This introduces a problem into the model in that the errors (in a statistical sense) are possibly correlated between observations. This affects the standard errors estimated from the model. It is unlikely to cause a major shift in standard errors, given that there are a large number of airports involved. While there are repeated observations at the same airport, they are not so common (relatively) as to dominate the estimation sample. Future research into airport models could attempt to account for repeated observations at the same airport via clustering or another method. The results of the ordered model are presented below. This model does not satisfy the assumptions of the ordered model (as seen by the Ordered Test P-value in Table 188). However, when category D incursions are excluded (as seen in Table 189), the model does conform to the assumptions of the ordered model. This supports the idea that category D incursions follow a separate process from categories A through C and may not be part of the same continuous ordering. Table – Ordered Logit Results for Airport Variables
Table – Ordered Logit Results for Airport Variables, Conflict Only
Although the overall model is invalid because of the ordering assumption, it is still worth noting some of the results. First, number of runway intersections plays a role. When excluding category D, this variable’s coefficient is both larger and considered more significant (but is less precisely estimated). This same situation can be seen for overall runway count, although the effect is in the opposite direction, reducing severity. The number of hotspots at an airport is only (marginally) significant when category D incursions are included. The expectation is that this variable may help explain category D in the multinomial model, and no other categories. A similar expectation is held for daily operations, which serves as an overall control on the frequency of incursions (i.e., incursions are more likely with more traffic, even if the rate of incursions per operations is constant), yet is no longer significant when category D incursions are excluded. Thus, daily operations may help explain category D but not the other categories. As discussed above, a simpler alternative to the multinomial model is to combine categories C and D and categories A and B. While ultimately a loss of detail, these models are simpler to interpret and focus the discussion on the impact on severe incursions – the categories of most interest for preventing crashes. The results of this binary logit are presented in Table 190. Table – Binary Logit Results for Airport Variables
The results for the binary logit are not dissimilar to those for the ordered model with all four severity categories. As in the ordered model, number of runway intersections increases the likelihood of a severe event. The impact is actually comparable in size to the impact in the ordered model, though these are expressed as odds ratios: for each additional runway intersection, the odds of a severe incursion are increased by approximately 25%. Counteracting this is the impact of having additional runways, which reduces the odds of a severe incursion by approximately 30% for each additional runway. Exposure (i.e., total operations) also plays a role in increasing severity, as seen in the ordered model; however, its impact is marginal at best. The results from the multinomial logit support many of the conclusions drawn above. There are no categorical variables, so the impacts of all variables are depicted in the following charts. As in the ordered and binary models, increasing numbers of runway intersections are associated with increased severity. This change in probability appears to result from a decrease in the probability of category C incursions. This may suggest that runway intersections are associated with conflict events rather than category D incursions. Table – Multinomial Logit Results for Airport Variables
Table – Results of IIA Test for Airport Variables
Figure – Impact on Probability of Severity Categories of Number of Runway Intersections The effect of number of runways appears, on the other hand, to be mostly a shift from the severe categories to category D. One potential explanation is that increased alternative runways can reduce the number of operations that could conceivably conflict. The impact of this variable is also fairly dramatic across the range seen in the dataset.
Figure – Impact on Probability of Severity Categories of Number of Runways Number of hotspots presents an interesting effect. The only severity category that appears to change over the range of this variable is category B. Overall, the impact of this variable appears to be to reduce severity – both categories C and D to increase in area on the chart. However, the impact on category B is still surprising.
Figure – Impact on Probability of Severity Categories of Number of Hotspots Daily operations also have a fairly strong impact. The impact is consistent with that seen in the ordered and binary models as well as models containing other sets of variables. The increased severity is likely explained by the increased probability of a conflict event, although the relative probability of category A incursions increases over the range of the variable.
Figure – Impact on Probability of Severity Categories of Daily Operations, Airports 
Figure – Impact on Probability of Severity Categories of Percent of AC/AT Traffic 
Figure – Impact on Probability of Severity Categories of Difference between Percent AC/AT and GA Traffic
|
