U. S. Department of Transportation
Download 2.66 Mb.
|
RadarThese models encompass the various radar technologies available in the dataset. This allows for a comparison between the various systems and their impacts on severity. It is important to note that the ASDE flag in this mode represents any form of ASDE; no distinction was made in the Runway Incursion dataset between ASDE-3 (or earlier) and ASDE-X. Additionally, ARTS was into simplified into variables representing their major version numbers (II or III). Table – Ordered Logit Results for Radar Variables
As these are a series of binary flags, it is important to remember that the alternative to these variables is that the respective system is not in place. Neither ARTS nor ASDE appears to reduce incident severity for OE incidents. STARS, on the other hand, appears to provide some benefit in terms of reducing severity. Interestingly, the interaction between STARS and ASDE is significant at approximately the 6% level. This is inconsistent with the results seen in Table 113, which indicated the interaction effect was insignificant. Additionally, the evidence of the benefit of ASDE seen in Table 111 is no longer observed, likely due to the inclusion of daily operations, which is highly correlated with ASDE (correlation = 0.58). Overall, the model with four severity alternatives does not satisfy the ordering constraint, indicating that these results are not indicative of the true relationship between these variables and severity. When excluding category D events, the ordering constraint is met, but no variable is significant. Table – Ordered Logit Results for Radar Variables, Conflict Only
The binary results are similar to the ordered model. Interestingly, nothing is significant at the standard five percent level (STARS is significant at a 7% level and is the only variable significant at a reasonable level). Table – Binary Logit Results for Radar Variables
Table – Multinomial Logit Results for Radar Variables
Table – IIA Test Results for Radar Variables
The multinomial model does little to clarify the results. Additionally, note that this model does not satisfy the IIA assumption when category B incursions are excluded. Although these tests are not particularly powerful, it is important to acknowledge that this model might violate that assumption in some cases. Because this is a series of flags with interactions, the predicted probabilities for each category are depicted in Table 198. The baseline airport has ARTS II. Table – Predicted Probabilities for Different Radar Combinations
Figure 54 depicts the impact of daily operations on severity categories. The only category for which this variable is significant is category D. As seen in other models containing this variable, increased daily operations are associated with increased severity. 
Figure – Impact on Probability of Severity Categories of Daily Operations, Radar It appears that, as seen in the ordered and binary models, STARS reduces the likelihood of severe incidents; however, this appears to be mostly a reduction in category C. Adding ASDE to STARS actually increases the likelihood of category C compared to only STARS, but ASDE alone also reduces the of likelihood of conflict incidents. As mentioned previously, it is possible that these effects are capturing the distribution of radar among airports. That is, ASDE is may be deployed at mostly busier airports that are more likely to have conflict events (due to the higher traffic). Additionally, the model coefficients are not precisely estimated – even when they are statistically different from zero. Thus, this model suggests that STARS may have some benefit in terms of reducing severity, but the results on other radar systems are inconclusive and provide little information beyond that provided by the categorical tests presented in Section 3.3.4.
|