Development of a Speeding-Related Crash Typology
PDF files can be viewed with the Acrobat® Reader®
SUMMARY AND CONCLUSIONS
This report described a large set of analyses aimed at developing information for use in better defining new treatments that could reduce SR crashes or crash severity and in better targeting existing treatments. It is difficult to summarize the findings of these analyses due to the number of outputs produced and the fact that the outputs were from two different methodologies–analysis of single–variable tables and regression tree development. It is also difficult because four different databases were used. General findings based on the overall results include the following:
There were differences between the fatal and total crash data, between the State and national crash data, and between the States. The differences
between factors related to fatal and total (i.e., primarily nonfatal) crashes in the FARS and GES data were not unexpected since they were paralleled by
similar difference in many other studies–fatal SR crashes differed to some extent from nonfatal crashes. This must be kept in mind during treatment
development and may sometimes lead to a choice between whether to attempt to affect fatal or total crashes. The expected differences between the State and
national data and the States only indicated that while speed reduction treatments developed for the United States as a whole should clearly be applicable to
all or many of the States, the priorities for the use of different treatments and the targeting of the treatments may change from State to State, reinforcing
the need for each State to develop a data–driven plan, namely their strategic highway safety plans. Few differences were seen between the results based on the two definitions. The use of the two definitions for SR–the combined definition
including over the posted speed limit and too fast for conditions and the over speed limit definition–did not reveal many changes in findings when used
in the two States. The somewhat expected difference between weather and road condition variables was found (e.g., dry conditions having more SR crashes in the
over speed limit definition and the opposing nondry conditions being most important in the combined definition), but most other findings were somewhat
consistent under both definitions. This lead to supporting the national finding in which only the combined definition was possible. The CART findings were at times more difficult to interpret than expected. The CART regression tree analyses provided information on interactions
between variables and on defined subsets of the data that could be targeted for future efforts. However, there were times when allowing CART to choose the
combinations of categories that were high SR and low SR within a given variable led to some combinations that did not provide clear information (e.g., the
combination of no restraint and lap and shoulder use in the higher SR branch of the GES vehicle–based tree). In addition, even though the full analyses
were restricted to only four levels of branches, there were times when the most important subcategory in terms of percentage of SR crashes included only a
small percentage of either total crashes or total SR crashes. CART is not a perfect methodology; it requires careful interpretations and usage.
With respect to more specific findings concerning high–priority variables and categories, in general, the findings were consistent with the findings from the study by Bowie and Walz and somewhat consistent with the 2001 study by Hendricks, et al.(5,8) Higher SR percentages in single–vehicle crashes, rural crashes, crashes on curves, nighttime crashes, motorcycle crashes as well as crashes involving male drivers, drivers not using restraints, and drivers using alcohol were found in either one or both of those earlier studies and in these current findings. In addition, key findings (including some which were inconsistent) from the current single–variable analyses include the following:
Younger drivers were more likely to be involved in SR crashes including drivers who were 21–25 years old. Regardless of database or definition,
the percentage of SR crashes was highest for the youngest drivers (16–19 years old) and decreased with age. It should be noted, however, that the 20
–25–year–old age group was consistently higher than the older categories. In fatal crashes, drivers with prior speeding convictions were more likely to be involved in SR crashes. Based on FARS data, the percentage of SR
crashes increased with prior speeding convictions. For drivers with three or more prior convictions, the SR percentage was almost twice that of those with
no prior convictions (32.4 percent versus 18.9 percent). Speeding did not seem to be as important a factor in crashes involving pedestrians and bicycles. All databases and definitions indicated that the
pedestrian and bicycle crashes were less likely to be SR than other type crashes. The nonpedestrian/bicycle crashes had two to five times the SR percentage as
the crashes involving pedestrians/bicycles. This could result partly from the fact that, even at low speeds, pedestrians and bicyclists were likely to be
injured. For SR crashes, high–priority, functional classes differed by database. While findings from GES and the two States showed agreement using the
combined definition for SR crashes (i.e., interstates, particularly rural interstates in the State data, had higher SR percentages) the FARS results differed,
showing minor collectors and local roads as having higher SR percentages. The use of the over speed limit definition in the State data provided even more
complexity with North Carolina data, showing that speeding over the posted limit was a bigger problem on local and minor roads. However, Ohio data showed a
bigger problem on interstates and arterials. These same trends in findings were reflected in the analyses of speed limits and number of lanes. GES and the
State data using the combined definition indicated that the SR crash percentages increased with speed limit and the number of lanes. FARS data indicated
roadways with lower speed limits and two lanes have higher SR crash percentages. The conclusion is that fatal crashes differed from total crashes, and crashes
likely differed between States. For SR crashes, high–priority, functional classes differed by database. While findings from GES and the two States showed agreement using the
combined definition for SR crashes (i.e., interstates, particularly rural interstates in the State data, had higher SR percentages) the FARS results differed,
showing minor collectors and local roads as having higher SR percentages. The use of the over speed limit definition in the State data provided even more
complexity with North Carolina data, showing that speeding over the posted limit was a bigger problem on local and minor roads. However, Ohio data showed a
bigger problem on interstates and arterials. These same trends in findings were reflected in the analyses of speed limits and number of lanes. GES and the
State data using the combined definition indicated that the SR crash percentages increased with speed limit and the number of lanes. FARS data indicated
roadways with lower speed limits and two lanes have higher SR crash percentages. The conclusion is that fatal crashes differed from total crashes, and
crashes likely differed between States.
With respect to the CART analyses, there were inconsistencies with the findings from the different databases as follows:
CART crash–based findings consistently identified single–vehicle crashes during adverse weather as high–priority subgroups. GES
included rear–end crashes as a top priority target. While there were no other significant findings for the combined definition in the GES and North
Carolina data, FARS added curved roads with lower speed limits to the key descriptors of fatal SR crashes. Analyses of the North Carolina and Ohio data using
the over speed limit definition indicated a key difference from the combined definition findings in that dry (or dry and wet) conditions replaced the snowy/icy
conditions as key descriptors. This implied that roadway–based treatments such as improved snow and icy removal and speeding–enforcement treatments
might both be appropriate for treating SR single–vehicle crashes. CART vehicle–based findings indicated that there was almost no consistency across databases, with perhaps young male showing up more than other
descriptors. FARS data indicated that in fatal crashes, alcohol use was the primary predictor, with drivers being young and male as further descriptors. In
the total crash datasets using the combined definition, GES noted distraction as the only important vehicle–based predictor, while the North Carolina and
Ohio databases indicated young males as the primary intervention targets. The North Carolina and Ohio data produced completely different results using the
over speed limit definition. North Carolina data indicated that the most important target subgroup was drinking drivers up to age 35 who were not using
restraints, and Ohio data indicated 16–25–year–old males who were not using restraints as the most important target subgroup.
This study produced a large group of findings which were not all consistent across the four databases and two definitions. This was not totally unexpected, as prior studies of other crash types not related to speeding have shown that the characteristics of fatal and nonfatal crashes do differ, and States would be expected to differ from each other at times and from a composite national picture. This effort produced some consistent (and inconsistent) findings that can be used in target development and targeting. The findings were consistent with those from the SR study conducted in 1994 by Bowie and Walz.(4) In essence, the problem characteristics have not changed much, and the problem is still a significant one that demands attention. The current focus on the issue is well justified and of critical importance in further reducing the cost to society resulting from motor–vehicle crashes in the United States.
FHWA-HRT-10-024
|
