U. S. Department of Transportation
Severity Research on Other Modes
Download 2.66 Mb.
|
- Navigate this page:
- Safety Research
- Methodological Concerns
Severity Research on Other ModesWhile research focusing on incursion severity seems to be lacking from the current runway incursion literature, the question of factors contributing to automobile crash severity has been examined extensively. This highway literature can provide important insight into how to approach modeling runway incursion severity. In addition, reviewing crash severity literature can illuminate those areas were runway incursions are similar to and diverge from the highway crash literature and will require careful consideration.
Schneider IV et al. examined the factors contributing to driver injury severity along horizontal curves in Texas.18 A multinomial logit approach was used and separate models were developed for three different curve radii (small, medium and large). Some of their findings can be translated to a runway incursion framework while others are less easily translated. The authors found that not wearing a seatbelt greatly increased the chance of a fatality. The same is true for the presence of alcohol and drugs. Those factors have no clear analogues in the runway incursion framework. The authors also examined environmental factors and found that clear weather and daylight increase the chance of a less severe accident. Weather may also play a role in runway incursion severity. Another factor the authors considered was vehicle type. Certain vehicle types (motorcycles) were associated with higher probabilities of more severe injuries while others (semi- and pickup trucks) were not. This translates rather directly into examining the impact of aircraft type on the runway incursion severity. However, the relationship between pilot experience and aircraft type would need to be carefully considered. Kockelman and Kweon also examined the factors contributing to driver injury severity.19 The authors used an ordered probit methodology and focused on different types of crashes: single versus two vehicle crashes. Again, the authors found a relationship between driver injury severity and vehicle type as well as alcohol. Interestingly, the authors did not find an effect for daylight (versus nighttime) on injury severity. The authors also found evidence of a non-linear relationship between injury severity and driver age. It is unclear how age may translate into a useful concept for runway incursions, but it speaks to the need to examine the included variables in a non-linear way as well. Lastly, the authors examined how the angle of the crash – head-on versus rear-end for example – contributes to driver injury severity. This suggests examining a similar notion of angle for runway incursions. For example, it may be that more severe incursions are associated with more certain relative angles between aircraft.20 Islam and Mannering provide another example of a multinomial logit approach.21 The authors focused on differing gender-age group combinations (male and female, young, middle-aged, and elderly drivers). This paper examines automobile-specific factors that could have contributed to injury severity. However, coefficients are reported for only some of the models (and then only the statistically significant ones), and select elasticities are reported in the comparison tables. This makes it difficult for the reader to gain a full understanding of implications of the model and removes the context for the results. Additionally, findings that are not statistically significant are as important as those results which are statistically significant. Reporting even insignificant results is a critical step in the research process. This analysis does provide an interesting template for comparing different subgroups of a population. Lam provides another example of an ordered probit approach targeted at comparing different age groups in a graduated licensing system in Australia.22
Xie et al. used a similar ordered probit model but the coefficients were estimated using a Bayesian approach.23 They examined the outcome of using different priors on the coefficient estimates. They also compared the results of standard ordered probit to a Bayesian ordered probit on the complete and a restricted sample to gauge the impact the differing methodologies had when compared on a small sample of data, a property of interest for statistical models. The restricted sample represents a random selection of 100 records from the complete set of 76,994 records. In the complete sample, they found results consistent with other studies: increased age and alcohol usage increase the injury severity. Both being male and certain vehicle types (vans and SUVs) reduce injury severity. The researchers found similar results between the standard ordered probit and Bayesian ordered probit in terms of coefficient magnitudes and standard errors for the full sample. When examining the restricted sample, the authors found that the Bayesian ordered probit provided answers more similar to those obtained on the full sample. This indicates that the Bayesian approach may be better suited to examining small datasets. Abdel-Aty used an ordered probit approach and found similar results when looking at crashes at three different roadway types in Florida (roadway sections, signalized intersections, and toll plazas).24 The author also tested these results against differing estimation procedures. Ordered logit models gave similar results, while a multinomial logit did not perform as well (as measured by how well the model predicted the known data and with fewer variables found to be significant). A nested logit procedure was also tested, but was found to be difficult to implement; the model also provided little improvement over the ordered probit in terms of model fit. The analysis provides insight into some methodological considerations but is not as informative for examining runway incursions. The variables used are specific to the road sections considered (such as whether or not an electronic toll tag was in use). Perera and Dissanayake also used an ordered probit approach.25 Their analysis focused on injury severity among older drivers. They developed two models, one for urban roads and one for rural. They found similar results as other studies, however the analysis is simplistic. For example, they used a series of binary variables to represent vehicle type. The general form of the variables is that they are equal to one if the vehicle was that type, and zero otherwise. They included binary variables for cars, vans, pick-ups, and SUVs. Note that these categories are by definition mutually exclusive: a car cannot be a van or a pickup or an SUV – knowing that one of the variables is equal to one reveals the value of the other vehicle variables. All coefficients for these variables are positive in the rural model. The authors report that the vehicles are associated with increased injury severity. However, without a reference case, the positive coefficients are inherently meaningless and must be compared amongst themselves. Pickups, with the lowest positive coefficient, thus reduce injury severity compared to other vehicle types rather than increase injury severity. The focus on older drivers and driver age renders this paper not very informative for runway incursions. However, it is illustrative of a methodological trap that needs to be avoided. These papers present a summary of the types of methodologies that may be used to understand runway incursion severity. Yet, the papers have some flaws worth noting with the intention that the same flaws are avoided during the modeling process for the current research. Several of the papers suffered from reporting deficiencies, such as not reporting all coefficients. Other papers suffered from methodological problems in their variable definitions or interpretation, such as the Perera and Dissanayake paper just described. While this research is suggestive of methodologies and factors to consider for runway incursions, there is a subtle difference between crash injury severity and runway incursion severity. Crash injury severities are conditional on a crash having already occurred whereas runway incursions are attempting to classify the underlying risk associated with an incident. It is important to keep these differences in mind when using injury severity literature to inform a study on runway incursions. While the underlying methodology will not change, the interpretation of the coefficients will be slightly different.
|