CHAPTER 5. FINAL ANALYSIS
5.2 Preliminary Analysis
5.2.1 Categorical variable analysis
5.2.1.1 Driver-related variables
The most important variables considered here are the driver injury severity, the driver’s age and gender. The first is the target of our analysis, the latter are the most significant effects found in the literature. The injury severity levels by driver section are considered in Table 5-5 below includes all the crashes on state multilane high speed roads. Sampling drivers from sections 1 and 2 resulted in a proportion of driver injury severity similar to that of the vehicle-driver sections 1 to 4. Records with missing and invalid data (from any of the variables) were removed. In the case of injury severity alone, removed records (injury severity levels 0 and 6) accounted for 12.15% of the total driver 1 records and 3.66% of the driver 2 records.
Table 5-5 Total frequency of involvements by injury severity for different driver sections
Driver sections 1-4 Driver 1 Driver 2 Injury Severity
Level Frequency Percent Frequency Percent Frequency Percent 0 37130 7.98% 26333 12.11% 7359 3.65%
1 254424 54.69% 118957 54.71% 104714 51.91%
2 90812 19.52% 33488 15.40% 49559 24.57%
3 57579 12.38% 26135 12.02% 28515 14.14%
4 23241 5.00% 11129 5.12% 11019 5.46%
5 1903 0.41% 1326 0.61% 543 0.27%
6 94 0.02% 81 0.04% 12 0.01%
Total 465183 100.00% 217449 100.00% 201721 100.00%
The complete data showed in the right portion of Table 5-1 corresponded to 120,421 crashes during the years 2002-2004. Two separate datasets were created for the preliminary analysis. The first consisted of 120,421 involvements of the first driver in a multiple or single vehicle crash. The second dataset of driver involvements (not from single vehicle crashes) had more complete records (n2=127,819) because it was sampled independently from the driver section 1 dataset for the preliminary analysis. Crashes with complete information in the driver 2 section were included in the second dataset, even when the driver 1 section was incomplete.
Some trends of single and multiple crashes can be identified in the tables discussed below.
Table 5-6 Driver Age group by injury severity for vehicle-driver sections 1 and 2
Driver-vehicle section 1 Driver-vehicle section 2 Driver Age
Percent 93.59% 6.41% 120421 100.00%
Test of independence p-value=<.0001,
Percent 94.34% 5.66% 127819 100.00%
Test of independence p-value=<.0001, Contingency Coefficien =0.0158 t
The age of the driver was a contributing factor in the exploratory models presented in Section 4.2. Middle age drivers (25-64 years old) compose the majority of the involved drivers in both sections, driver section 2 (69.91%) somewhat higher than section 1 (61.29%). The major
difference in the proportions is for the very young drivers (15-19 years old), which represent 12.52% of the involved drivers in section 1 versus only 8.62% in section 2. This suggests an increased involvement of youngsters in single vehicle crashes. Clearly, their proportions of severe injuries are lower than for any of the age groups, as their physical condition is generally most favorable in case of a crash. For multilane high-speed arterials, there is a significant difference in the proportion of severe crashes for older drivers, as shown in Table 5-6 above.
When comparing the driver 1 and driver 2 sections, the effect of single vehicle (off-road) crashes is perceived as increasing the chance of severe crashes. This was investigated in the final analysis models by testing the interaction variable driver age and off-roadway crash.
Land use (as a surrogate of travel choice) may also influence the severe crash outcomes for different driver age groups, an interaction (age group and rural/urban) was also tested in the final analysis. Another set of interacting variables (driver age and driver at-fault) was tested to determine if the decreased severity odds ratio found in the exploratory analysis holds for all age groups, but it caused numerical problems (quasi-separation) explained in section 5.4.2. This result tends to confirm the theory of the driver at-fault bias in the driver injury severity modeling.
Table 5-7 Driver Gender by injury severity for vehicle-driver sections 1 and 2
Driver-vehicle section 1 Driver-vehicle section 2 Gender1 Severe driver1
Percent 93.59% 6.41% 120421 100.00%
Test of independence p-value=0.0001,
Percent 94.34% 5.66% 127819 100.00%
Test of independence p-value=<.0001, Contingency Coefficient=0.0149
Gender was a variable of major importance in all of the exploratory models, as shown in section 4.2.2. Different driver behavior and physiological characteristics play a role in the different outcomes of a crash event. Table 5-7 above shows that females have a larger proportion of severe injuries, while males have a larger number of severe injury involvements. There seems to be an overrepresentation of male total and female severe involvements, when compared to the general population. However, there is no direct gender exposure measure of the driving population on arterial corridors.
Table 5-8 Safety equipment used by injury severity for vehicle-driver sections 1 and 2
Driver-vehicle section 1 Driver-vehicle section 2 Safety
Percent 93.59% 6.41% 120421 100.00%
Test of independence p-value=<.0001,
Percent 94.34% 5.66% 127819 100.00%
Test of independence p-value=<.0001, Contingency Coefficient 0.1569 =
The use of seat belts has been an important factor in reduced injury severity odds ratio for the exploratory analysis. As shown in Table 5-8 above, the rate of severe injuries is much higher when no seat belt use is reported. However, previous studies (such as Richardson, et al., 1996) have found seat belt use over reporting for the non-severe crashes to avoid traffic fines. For severe crashes, it is usually possible for the police officer (or EMT) to determine if the injured occupant was using a seat belt.
The rate of seat belt use for non-severe injuries is quite high (between 87 and 92% for driver 1 and 2, respectively), while for the severe injuries is between 61 and 76%. The difference in the rates for the severe and non-severe crashes is between 17 and 28%. The official rate of usage across Florida in 2004 was 76.3% (FDOT, 2008). Comparing the rate of seat belt use of the non-severe crashes with the average use in Florida reveals that the over reporting could be as high as 13% for driver 1 and 17% for driver 2. Previous studies have suggested various rates of over-reporting of seat belt usage in non-severe crashes. A study by Streff and Wagenaar (1989) compared self-reporting of seat belt use to observational surveys of the same population. The authors’ best estimate was to discount self-reported rates by 12 percent. In a study of police-reported crash data in Hawaii, Li et al. (1999) found a 10% reduction in the reported seat-belt use rate when adjusting for over-reporting. In addition, Hawaii hospital data showed that physicians reported 63.59% seat belt usage, while police reports usage rate was 90.26%. These figures are somewhat similar to the percentages presented in Table 5-8 above, which suggests that there is an over-reporting of the seat belt usage in the Florida crash data.
Table 5-9 At-fault driver by injury severity for vehicle-driver sections 1 and 2
Driver-vehicle section 1 Driver-vehicle section 2 At Fault
Percent 93.59% 6.41% 120421 100.00%
Test of independence p-value=<.0001,
Percent 94.34% 5.66% 127819 100.00%
Test of independence p-value=<.0001, Contingency Coefficien =-0.0193 t
The driver involved in a crash cited for a moving violation is considered at-fault. This measure of legal responsibility also reflects unsafe driving behavior. The lower rates of severe driver injury for those found at-fault shown in Table 5-9 above agree with the results of the exploratory analysis. It is important to note that the drivers in section 1 have a stronger tendency to be at fault; however there is no evidence of a systematic bias for multiple vehicle crashes, as shown in section 5.3. On the other hand, a large proportion of driver 1 “innocent” drivers sustained severe injuries (9.45%), which is one of the highest proportions seen so far in this investigation. At issue is whether single vehicle crashes with severe crash outcomes are less likely to involve “unsafe” driver behavior.
A study of Central Florida signalized intersections by Abdel-Aty (2003) found a significant negative effect of drivers not at fault, possibly due to the driver at fault being the striking vehicle, which for angle and turning crashes is expected to experience a lower level of injury than that of the driver of the stricken vehicle. This factor was found significant in all the exploratory models, which may have broader implications for the multilane high speed arterials.
The speeding and contributing cause variables discussed next exhibit a similar situation with significant differences between the relationships for drivers 1 and 2. It is possible that police officers would tend to record cited drivers first and this compounded with the single vehicle crashes contributes to the higher proportion of severe injuries for the driver 1 section.
Throughout this analysis, there is evidence of important differences between the driver 1 and 2 sections and serve as investigative support for the sampling of drivers from sections 1 and 2 for the final analysis. Only one driver involvement per crash was analyzed, and stratified sampling proved to be a sound method, as discussed in section 5.3. This categorical data analysis served to indicate the association of these variables with the driver injury severity and to point out some
trends that were ultimately confirmed in the final analysis. The tests for independence had the same conclusion for both sections in most cases, as shown in Appendix B. Only in one case were the variable was clearly significant in this test for both sections were excluded from further analysis. Part of the validation of the sampling technique involves comparing the final models with the trends presented in this section to confirm whether the sample captured the effects form both driver sections. These sections represent a vast majority of the total driver involvements.
Table 5-10 Driver speeding by injury severity for vehicle-driver sections 1 and 2
Driver-vehicle section 1 Driver-vehicle section 2 Speeding1 Severe driver1
Percent 94.34% 5.66% 127819 100.00%
Test of independence p-value=<.0001, Contingency Coefficien =0.0678 t
Speeding is suspected to be a major factor in severe crashes. Table 5-10 above shows that the rate of severe crashes is less for those drivers found speeding than for those not speeding.
This result is similar to the at-fault driver results shown previously. The driver with the speeding citation would be at-fault and since it is usually the striking vehicle, it would cause severe damage to the stricken vehicle (usually not speeding). An interaction variable of speeding and point of impact was tested in the models to prove whether this theory was true. In addition, the proportion of non-speeding drivers in section 1 with severe injuries is of great concern.
Since the speeding indicator is computed using the estimated speed reported by the police officer only for certain types of crashes, there is more missing data than for any other variable. It is more likely that the police officer reports an estimated speed for a severe crash requiring a thorough investigation. Thus, it was deemed pertinent to include this variable (with one level labeled unknown) for its perceived significance in severe crash outcomes. The bias implications of this variable will be further discussed in the coming sections.
Table 5-11 Driver ejection by injury severity for vehicle-driver sections 1 and 2
Driver-vehicle section 1 Driver-vehicle section 2 Ejected1 Severe driver1
Percent 93.59% 6.41% 120421 100.00%
Test of independence p-value=<.0001,
Percent 94.34% 5.66% 127819 100.00%
Test of independence p-value=<.0001, Contingency Coefficien =0.1866 t
Ejection of the driver was found to be the most important variable affecting injury severity in the exploratory models. However, this is considered a post-crash event and in the exploratory models showed a propensity to cause numerical (quasi-separation) problems in some cases. This factor has been extensively considered in injury severity analysis on the interest of predicting its occurrence or effect on injury severity. As shown in Table 5-11 above, almost half of those drivers ejected in section 1 suffered severe injury. Given that the steering wheel turns into a source of injury, drivers without seat belts are expected to sustain higher degrees of injury.
Driver ejection has been primarily associated as a consequence of seat belt non-usage. However, it is not an exclusive determinant of severe injury.
As suggested by the reduced severe injury percentage for the drivers in section 2, driver ejection has been considered as an important part of the sequence of events in a roadside crash.
Possible interactions are tested in the final analysis to investigate how this outcome is related to some crash precursors.
Table 5-12 Driver contributing cause by injury severity for vehicle-driver sections 1 and 2
Driver-vehicle section 1 Driver-vehicle section 2 Contributing
driver action (96.06) (3.94) 15869 13.18%
32847 2484
Percent 93.59% 6.41% 120421 100.00%
Test of independence p-value=<.0001,
driver action (94.35) (5.65) 110363 86.34%
4900 275
Percent 94.34% 5.66% 127819 100.00%
Test of independence p-value=<.0001, Contingency Coefficien =0.0245 t
This parameter represents the most common driver actions that contribute to a crash, as reported by the police officer. The major categories used in this investigation were aggressive driving and alcohol/drug influence. Other categories had sparse data and would not provide practical results in the models used in the final analysis. As shown in Table 5-12 above aggressive driving is an important factor in severe crashes, especially for the driver 1 section;
18.45% of the severe injuries driver involvements involved aggressive driving. However, when
we compare the 29.34% aggressive driving involvements in section 1 with 4.05% in driver section 2, there is a possible bias in driver section a towards innocent (no improper action) drivers. The actions constituting aggressive driving include speeding, failed to yield right-of-way, improper lane change, followed too closely, improper passing & disregarded other traffic control. The implications of the differences found between drivers sections 1 and 2 are further discussed in section 5.3.
Table 5-13 Driver physical defects by injury severity for vehicle-driver sections 1 and 2
Driver-vehicle section 1 Driver-vehicle section 2 Physical
Percent 93.59% 6.41% 120421 100.00%
Test of independence p-value=<.0001,
Percent 94.34% 5.66% 127819 100.00%
Test of independence p-value=0.963, Contingency Coefficien =-0.0001 t
The physical condition of a driver is expected to be a contributing factor in crashes. It was tested in the final analysis to find out whether this uncommon occurrence does have a significant impact in driver injury severity. The proportion of severe driver injury severity for section 1, as shown in Table 5-13 above, is more than 12%, which is the highest for any variable reviewed so far. Most of the physiological conditions listed as defects are related to sight, hearing and fatigue. However, in the preliminary analysis other conditions such as, seizure, epilepsy or blackout registered above 20% severe injuries, the largest contributor to severe injuries due to the physical defects.