Secondary Collisions Following a Traffic Barrier Impact: Frequency, Factors, and Occupant Risk

Abstract

This study has investigated secondary collisions following an initial barrier impact in tow-away level crashes. The analysis included 2026 barrier impact cases that were selected from 12-years of in-depth crash data available through the National Automotive Sampling System (NASS) / Crashworthiness Data System (CDS). Secondary collisions were found to occur in approximately one-third of tow-away level crashes where a traffic barrier was the first object struck. Secondary crashes were found to primarily involve an impact to another vehicle, an impact to another barrier, or a rollover; tree and pole impacts were found to represent a much smaller proportion of secondary impacts. Through a detailed analysis of vehicle trajectory, this study supports previous research suggesting secondary collision risk is substantial even for vehicles not ultimately involved in a secondary collision. Compared to a single barrier impact, the occurrence of a secondary collision was found to increase the risk of serious occupant injury by a factor of 3.5, equivalent to the serious injury risk difference found between a belted and unbelted occupant in a traffic barrier crash.

INTRODUCTION

A longitudinal traffic barrier redirects an impacting vehicle in order to prevent it from colliding with a more dangerous roadside object. Ideally, the impacting vehicle should come to rest alongside the barrier to minimize the risk of a secondary collision. Examples of a secondary collision would include, but are not limited to, a vehicle being redirected into an adjacent traffic lane and impacting another vehicle or a vehicle impacting a tree following the initial barrier impact. Previous real-world crash analysis has revealed that these secondary impacts are not uncommon and, when they occur, result in a greater risk of fatal and severe occupant injuries. Although full-scale crash test evaluation procedures attempt to address post-impact vehicle trajectory, it is rarely a discerning factor in the assessment of traffic barrier crash performance. Further, investigation of this issue has been limited to studies conducted over 20 years ago and only at the state, not the national, level.

OBJECTIVE

The purpose of this study is to investigate secondary collisions following an impact with a traffic barrier with more recent national level data to determine (1) the frequency of these crashes, (2) the resulting influence on occupant injury, and (3) how factors, such as barrier type and vehicle type, affect the occurrence and severity of these crashes.

BACKGROUND AND PREVIOUS RESEARCH

Longitudinal barriers, such as w-beam guardrails, must demonstrate adequate crash performance in a series of full-scale crash tests prior to being considered acceptable for use on the national highway system. Procedures for determining the crashworthiness of longitudinal barriers in the US are set forth in the Manual for Assessing Safety Hardware (AASHTO, 2009), which is a recent update to the procedures set forth in the National Cooperative Highway Research Program (NCHRP) Report 350 (Ross et al., 1993). Analogous European crash test procedures are prescribed in EN-1317 (CEN, 1998). These test procedures specify high speed oblique angle impacts into a barrier with both a passenger car and a large pickup truck. The intent is to evaluate barrier crash performance under practical worst-case conditions. The results of the tests are evaluated against specific criteria in three areas: (1) the structural adequacy of the barrier, (2) the potential for injury to vehicle occupants, and (3) the vehicle trajectory as a result of the impact.

The longstanding US criteria for post-impact vehicle trajectory was a recommendation that the vehicle exit the barrier at an angle less than 60 percent of the impact angle and that the vehicle not intrude into adjacent traffic lanes (Ross et al., 1993). Although quantitative, this recommendation was simply a “preferable” condition and not a strict requirement for a passing barrier crash test. It should be noted, however, that the recently updated criteria do require that a vehicle exit the barrier within an “exit box” of specified dimensions dependent on vehicle length and width (AASHTO, 2009).

A small number of previous studies have investigated the issue of post-impact vehicle trajectory using real-world crash data. Ray et al. (1986) examined 679 barrier crashes: 124 crashes from a Texas bridge rail study and 555 crashes from the Longitudinal Barrier Special Study (LBSS). Of the available data, only 17 cases, or less than 3 percent, were found to result in serious or severe occupant injury. More than 80 percent (14 of 17) of these crashes involved a secondary collision. From a detailed reconstruction of each of the cases with severe injury, the authors found that none of the occupants sustained the serious injury during the initial barrier impact. Also, in each case, the injury sustained in the secondary collision was found to be more severe than the injury level attributed to the initial barrier impact.

Ray et al. (1987a, 1987b) used fault tree analysis to further characterize increased injury risk in secondary collisions. Police-reported crash data was analyzed from two sources: 325 barrier crashes from North Carolina and over 3,000 single vehicle crashes from New York State. Based on the data from both sources, fatal and incapacitating injury was found to be 3 times as likely for crashes involving a secondary collision than crashes with no secondary collision. Further, available vehicle trajectory data suggested that redirection of a vehicle onto or across the roadway following a barrier impact was not an uncommon occurrence. More than 75 percent of barrier crashes in the data resulted in some level of intrusion into or across adjacent travel lanes.

There are two primary limitations with these existing studies. First and foremost is the age of the crash data. The LBSS data used by Ray et al. (1986) included crashes from 1982 through 1983; similarly, the data from Ray et al. (1987a, 1987b) included crashes occurring between 1980 and 1983. Now in excess of 20 years old, these data are no longer indicative of the current vehicle fleet or the current field-installed barriers. In addition, both studies were largely limited to state level data and not necessarily representative of barrier crashes across the US. These limitations coupled with the small number of studies provide a strong motivation for revisiting this issue.

METHODS

Data from the National Automotive Sampling System / Crashworthiness Data System (NASS/CDS) was used to examine secondary collisions after an impact with a traffic barrier. NASS/CDS provides a detailed record of approximately 5,000 crashes investigated each year (NCSA, 2005). To be included in NASS/CDS, at least one of the crash-involved vehicles had to be towed from the scene. The NASS/CDS database includes only crashes involving cars, light trucks, vans and sport utility vehicles (SUVs); heavy vehicles and motorcycles are not included as subject vehicles in the NASS/CDS database. The complex sampling strategy used to select cases for NASS/CDS oversamples certain types of crashes including fatal crashes, crashes involving hospitalized occupants, and crashes involving late model year vehicles among other factors (NCSA, 2005).

To permit nationally representative estimates to be computed, NASS/CDS provides weighting factors which account for this complex sampling scheme. These weights were applied in the analysis which follows. All statistical analyses were performed using the SAS V9.1.3 software package.

Case Selection and Database Development

Cases were selected from a 12-year NASS/CDS data set spanning 1997 to 2008, inclusive. Cases that were initially selected from NASS/CDS fell into one of two categories:

1. The first and only event was a vehicle impacting a longitudinal barrier.

2. A vehicle struck a longitudinal barrier and subsequently impacted another vehicle, object, or overturned.

Only passenger vehicles, light trucks, vans and SUVs were included; any heavy vehicles were excluded from the analysis. No restriction was placed on the vehicle model year. For the purpose of this study, a longitudinal barrier included concrete barriers, metal beam guardrails, and cable barriers. Longitudinal barriers in NASS/CDS are grouped into one of two categories: (1) concrete barriers, and (2) other barriers. The latter category includes all types of steel guardrail systems such as w-beam guardrails, box beam barriers, and cable barriers. For the purpose of this study, the “other” category will be referred to as metal barriers.

Crash scene photographs available online for the NASS/CDS years 1997 through 2008 were used to determine relevant barrier and redirection characteristics for each crash. Barrier type characteristics included verifying the concrete or “other” barrier classification made by the NASS investigator, further classifying the “other” barriers, and determining if the vehicle struck the end of the barrier or the length of need section (portion between the end terminals). Cases were also checked to ensure that only barriers specifically designed to redirect an impacting vehicle were included; any miscoded objects were excluded from further analysis. A more detailed description of the methodology used to determine barrier type and impact location for each crash is provided by Gabauer and Gabler (2009).

Using the available scene diagram, written crash description, and crash scene photographs, the following additional information was determined for each suitable crash-involved vehicle:

1. Vehicle trajectory. The vehicle trajectory following the initial barrier impact was classified into one of 6 categories: next to barrier (NB), into adjacent lane or lanes (L), across all lanes (AL), beyond barrier (BY), behind barrier (BB), or unknown (U).

2. Barrier Performance. For each crash, an assessment of the barrier crash performance for the initial barrier impact was made using the data available. Performance was classified into one of 3 categories: no penetration (N), penetration (Y), or unknown (U).

3. Second Event Presence. For each case, an independent determination was made regarding the presence of a secondary collision.

Vehicle trajectories coded as NB included all cases where the vehicle came to rest between the face of the barrier and the first adjacent lane, with no evidence of intrusion into that lane. For barriers offset a significant distance from the roadway, a vehicle would only have to remain in the median or roadside area to be coded as NB. Any vehicles intruding into one or more of the adjacent travel lanes were coded as L. To be coded as AL, a vehicle had to cross at least all of the same direction travel lanes. A code of BY was used to indicate vehicles that rode alongside a barrier until the barrier end and subsequently continued off the road beyond the barrier. A code of BB was used for all cases where the vehicle penetrated the barrier. This included any form of uncontrolled penetration such as breaking through the barrier, vaulting over the barrier, passing underneath the barrier, or riding atop the barrier. The barrier performance in any of these instances would be coded as penetration; barrier performance was coded as no penetration otherwise. Cases were coded as unknown when the trajectory and/or barrier performance characteristics could not be determined to a reasonable degree from the available information. Cases with either unknown trajectory or performance were excluded from further analysis.

In several cases, NASS investigators would code two barrier impact events for a single barrier redirection event, especially for cases where the vehicle front contacted the barrier followed immediately by an impact by the vehicle side. As this sequence is typical of redirectional crashes, it would not be considered a secondary collision. The independent determination of presence of a second collision was intended to discern these cases from true secondary collisions.

Bridge rails and transition sections were also excluded from the analysis. Although these are types of longitudinal barriers, there were too few cases to provide a meaningful analysis. Vehicle type was determined using the “Bodytype” variable in NASS/CDS. The NASS/CDS rollover variable was used to determine if the vehicle overturned in the crash.

Statistical Analysis and Model Development

Contingency table analysis was used to provide a preliminary investigation of the factors that may influence the occurrence of a secondary collision, including barrier type and vehicle type. For each vehicle type, the secondary collision proportion was computed along with the confidence limits of those estimates that account for the variation introduced by the complex sampling design of NASS/CDS. A similar analysis was undertaken for barrier type (concrete barrier and metal barrier). In each case, a contingency table analysis was used to determine if differences exist in secondary collision rates by vehicle type and barrier type.

To better characterize current secondary collisions, the distribution of object struck for the impact immediately following the initial barrier impact was tabulated. In addition, the distribution of vehicle trajectory was computed for two data subsets: (1) crashes involving a secondary collision, and (2) crashes without a secondary collision. Similar to the contingency table analysis above, the corresponding 95 percent confidence intervals on each estimate were also computed. The distribution of vehicle trajectory for crashes without a secondary collision was intended to estimate the potential for a secondary collision and provide a point of comparison with previous research.

Two binary logistic regression models were developed to further investigate factors related to the occurrence of secondary collisions as well as injury risk differences in these crashes. The first model predicted the occurrence of a secondary collision based on barrier type, vehicle type, and other confounding factors. The second model predicted the presence of serious occupant injury based on the presence of a secondary collision while controlling for other confounding factors. For the purpose of this study, serious occupant injury was defined as a maximum Abbreviated Injury Severity (AIS) score of 3 or above (AAAM, 2008)

For the secondary collision presence model, confounding factors included impact location relative to the barrier and barrier penetration. Both variables were dichotomous; the barrier impact location variable indicated whether the vehicle impacted the length of need or the end terminal of the barrier while the penetration variable indicated if the vehicle penetrated the barrier.

For the occupant injury model, confounding factors were barrier type, vehicle type, and occupant restraint usage. Barrier type was dichotomous and indicated an impact with a concrete or metal beam barrier. Based on the “bodytype” variable, vehicle type included four categories: (1) passenger cars, (2) pickup trucks, (3) utility vehicles, and (4) vans. Occupant restraint usage indicated either lap and shoulder belt usage or no belt usage for a specific occupant. Any occupants with other belt usage, e.g. lap only, shoulder only, other or unknown, were excluded from this portion of the analysis. Ejected occupants, occupants in vehicles that penetrated a barrier, and occupants in vehicles impacting end terminals were also excluded from this portion of the analysis. Occupants with unknown injury severity were also excluded from this portion of the analysis. Occupants coded as “injury, severity unknown” were checked against the “treatment” variable; any of the cases with a treatment variable indicating a fatality were recoded to an AIS 6 injury.

Both developed models accounted for the first level stratification and clustering within NASS/CDS through the use of the “surveylogistic” procedure available in SAS. Case stratification in NASS/CDS is based on vehicle tow status, occupant injury level, and hospitalization (NCSA, 2005). The first level clusters are represented by the primary sampling units (PSUs) located across the United States. A more detailed description of the NASS/CDS sampling design methodology can be found in the NASS/CDS Analytical User’s Manual (NCSA, 2005).

RESULTS

Data Characterization

Using the initial selection criteria, 2400 NASS/CDS cases were available for review. Approximately 11 percent of these available cases were not suitable for analysis and were excluded. This included 80 bridge rail impacts, 22 transition impacts, 87 instances where the barrier type could not be verified, and 85 instances where the struck barrier was not specifically designed to redirect a vehicle. Objects that fell into the latter category included concrete median planters, crash cushions, poles, curbs and highway noise barriers.

Of the 2126 remaining crashes, redirection was unknown in 52 cases, barrier performance was unknown in 4 cases, and a determination of second event presence was not possible in 2 cases. In addition, there were 6 cases excluded due to missing vehicle type information. Also, there were 40 cases where the barrier type was coded improperly by the NASS investigator. A total of 32 concrete barriers had been coded as “other barrier” while 8 metal beam barriers had been coded as “concrete barrier.” After exclusion of the unsuitable barrier types and correction of the miscoded barrier types, there were a total of 2062 cases suitable for analysis. The final vehicle data set is summarized in Table 1 and represents just over 1 million impacts into traffic barriers.

Table 1.

Vehicle-Barrier Crash Data Set [NASS/CDS 1997–2008, inclusive]

Variable	Raw Cases	Weighted
All	2062	1,004,678
Second Event
Yes	932	339,850
No	1130	664,828
Vehicle Trajectory
Next to Barrier (NB)	577	327,895
Into Lane or Lanes (L)	818	354,839
Across All Lanes (AL)	435	226,845
Beyond Barrier (BY)	47	57,583
Behind Barrier (BB)	185	37,516
Barrier Type
Concrete	1114	462,039
Metal	948	542,639
Strong Post W-beam	774	433,307
Thrie Beam	77	41,302
Weak Post W-beam	53	31,599
Box Beam	29	27,332
Cable	15	9,099
Component Struck
Length of Need	1783	904,254
End Terminal	279	100,424
Vehicle Type
Car	1413	703,659
Light Truck	649	301,019
SUV	321	130,655
Pickup Truck	250	140,350
Van	78	30,014

Secondary collisions occurred in approximately one third of the available cases (34 percent weighted; 95% CI = 28.2 – 39.4). A majority of the vehicles (86 percent unweighted, 90 percent weighted) initially struck the length of need portion of the barrier as opposed to a barrier end. There was an approximately equal split of concrete and metal barriers in the available data; 46 percent (weighted) concrete barriers and 54 percent metal barriers. Of the metal barriers, the vast majority (80 percent weighted) were strong post w-beam barriers with a much smaller proportion of strong post thrie beam barriers (8 percent), weak post w-beam barriers (6 percent), box beam barriers (5 percent), and cable barriers (2 percent). The vehicle trajectory following the initial barrier impact involved the vehicle intruding into one or more adjacent traffic lanes (35 percent), coming to rest next to the barrier (33 percent), or crossing all adjacent traffic lanes (23 percent). A much smaller proportion of vehicles traveled beyond the end of the barrier (6 percent) or penetrated and traveled behind the barrier (3.75 percent). The vehicle mix was approximately 70 percent (weighted) cars, 30 percent light trucks, which included pickup trucks, sport utility vehicles (SUVs), and vans.

The final vehicle data set was combined with the available NASS/CDS occupant data to develop the data set used in the occupant injury model. In the 2062 barrier crashes, there were a total of 3148 occupants. A total of 553 occupants (17 percent) had a restraint usage other than lap and shoulder belt or no restraint; these cases were excluded from the analysis. The vast majority of these cases (421) had missing belt usage. Of the 2595 remaining occupants, 44 were excluded (1.4 percent of all available occupants) due to missing occupant injury data or missing ejection status (or both). Of the 2551 occupants remaining, 125 were excluded (4 percent of all available occupants) due to unknown injury severity. A total of 169 partially or fully ejected occupants (5.4 percent of all available occupants) were then excluded from the data set. Of the remaining occupants, 190 occupants (6 percent of all available occupants) were in vehicles that penetrated the barrier and were excluded. A total of 160 of the remaining occupants (5 percent of available occupants) were in vehicles impacting an end terminal and were subsequently excluded.

With all the exclusions, the final occupant injury data set consisted of 1907 occupants. Note that 3 of the occupants in the data set were converted from an unknown injury severity to an AIS 6 based on the “treatment” variable indicating a fatality.

Secondary Collision Frequency, Type and Vehicle Trajectory

Table 2 and Table 3 and provide a summary of the contingency table analysis of secondary collisions by vehicle type and barrier type, respectively. The tables show the weighted second collision proportion by vehicle / barrier type along with the 95 percent confidence intervals of these estimates.

Table 2.

Barrier Secondary Collision Rates and 95% Confidence Intervals by Vehicle Type, [NASS/CDS 1997–2008, inclusive]

Vehicle Type	Second Event	Weighted Percent	95% CI on Percent
Car	No	70.5	64.8 – 76.2
	Yes	29.5	23.8 – 35.2
Pickup Truck	No	64.8	49.1 – 80.4
	Yes	35.2	19.6 – 50.9
SUV	No	46.9	32.1 – 61.7
	Yes	53.1	28.6 – 67.9
Van	No	56.1	30.5 – 81.6
	Yes	43.9	18.4 – 69.5

Table 3.

Barrier Secondary Collision Rates and 95% Confidence Intervals by Barrier Type, [NASS/CDS 1997–2008, inclusive]

Barrier Type	Second Event	Weighted Percent	95% CI on Percent
Concrete	No	65.7	58.1 – 73.5
	Yes	34.2	26.5 – 41.9
Metal	No	66.5	58.0 – 75.0
	Yes	33.5	25.0 – 42.0

Cars were found to have the lowest percentage of secondary collisions (29.5 percent) while SUVs were found to have the highest proportion (53.1 percent). The difference in secondary collision occurrence rates between vehicle types was found to be statistically significant based on the Rao-Scott modified likelihood ratio chi-squared test (p = 0.0434). By barrier type, however, there was virtually no difference in secondary collision rates (p = 0.9045).

Table 4 presents the object struck distribution, both unweighted and weighted, for the impact immediately following the initial barrier impact. Impacts to another vehicle, impacts to another barrier, and vehicle rollover combined represent more than 80 percent of the secondary collisions. Pole and tree impacts represent approximately 5 percent of secondary collisions.

Table 4.

Summary of Object Struck in Secondary Collisions

Object Struck (Secondary Collision)	Raw Cases	Weighted Cases	Weighted Percent
Rollover	351	97,615	29.9
Concrete Barrier	191	94,611	28.9
Metal Barrier	104	47,021	14.4
Vehicle	145	51,190	15.7
Pole/Tree	44	16,460	5.0
Embankment	14	*	*
Ditch/Culvert	6	*	*
Other	77	20,047	6.1
Total	932	326,944

^*Note: Weighted estimates not computed due to small raw case sample size.

Figure 1 shows the weighted distribution of vehicle trajectory for those vehicles involved in a secondary collision and those vehicles not involved in a secondary collision. The error bars shown on each estimate represent the 95 percent confidence interval on the estimate, which accounts for the uncertainty introduced by the NASS/CDS sampling design. Vehicles involved in a secondary collision exhibit much higher proportions of the trajectories involving intrusion into or across adjacent traffic lanes and lower proportions of vehicles coming to rest next to the barrier. Nearly 40 percent of crashes with no secondary collisions, however, involved a vehicle that intruded into one or more of the adjacent traffic lanes.

Figure 1.

Distribution and 95% Confidence Intervals for Vehicle Trajectory following a Traffic Barrier Impact by Presence of a Secondary Collision.

Factors Affecting Secondary Collisions

A binary logistic regression model was used to predict involvement of a vehicle in a secondary collision following an initial barrier impact. A summary of the binary logistic regression model parameters is shown in Table 5. For each parameter, the Wald Chi-Square statistic and associated p-value has been included as well as the C-statistic for the model. The C-statistic represents the area under the Receiver Operator Characteristic (ROC) curve and provides a single numerical value of how well the model distinguishes between the response variable, in this case, presence of a secondary collision.

Table 5.

Summary of Binary Logistic Regression Parameters: Secondary Collision Presence Model

Parameter	Wald χ2	P	C
Barrier Type	0.148	0.7004	0.62
Vehicle Type	9.418	0.0242
Impact Location	1.243	0.2649
Penetration	26.87	< 0.0001

The effects of vehicle type and barrier penetration were found to be statistically significant (p = 0.0242, and p < 0.0001, respectively). Neither barrier type nor impact location was found to have a statistically significant effect in the available data set. Interactions between the model parameters were also tested. None of the interaction effects were found to be statistically significant. With the exception of the interaction between impact location and vehicle type (p = 0.0727), the p values were at or above 0.29.

Table 6 shows the odds ratios for barrier type, vehicle type, impact location, and barrier penetration for the binary logistic regression model. For the dichotomous barrier type, impact location, and barrier penetration variables, the odds ratio shown was with respect to the other possible value (i.e. metal barriers, end terminal impact, and no barrier penetration). The vehicle type comparison group was passenger cars. The 95 percent confidence bounds on each odds ratio are also shown.

Table 6.

Odds Ratios and 95% Confidence Intervals for Barrier Secondary Collisions

Parameter	Value	Odds Ratio	95% CI
Barrier Type	Concrete	1.11	0.6 – 2.0
Vehicle Type	SUV	2.60	1.4 – 5.0
	Pickup Truck	1.23	0.6 – 2.5
	Van	1.86	0.6 – 5.6
Impact Location	Length of Need	1.55	0.7 – 3.3
Barrier Penetration	Yes	8.59	3.8 – 19

SUVs were found to be more than twice as likely to be involved in a secondary collision compared to passenger cars; this result was statistically significant. Also, vehicles that penetrate the barrier were found to be nearly nine times as likely to be involved in a second collision. This includes vehicles that hit and penetrate a barrier end terminal. Although not statistically significant, there was an increased propensity for involvement in a secondary collision for impacts with concrete barriers, length of need impacts, as well as impacts involving vans and pickup trucks.

Occupant Injury in Secondary Collisions

Results from the occupant injury binary logistic regression model are summarized in Table 7. The Wald Chi-Square statistic and associated p-value has been included for each parameter as well as the C-statistic for the entire model. The effects of secondary collision and restraint use were found to be statistically significant (p = 0.0191, and p = 0.0223, respectively). Neither barrier type nor vehicle type was found to have a statistically significant effect on serious occupant injury in the available data set. Interactions between the model parameters were also tested for the occupant injury model. With the exception of the interactions between presence of a second event and barrier type (p = 0.0407) and between barrier type and vehicle type (p = 0.0471), none of the interactions were found to be statistically significant.

Table 7.

Summary of Binary Logistic Regression Parameters: Injury Severity Model

Parameter	Wald χ2	P	C
Secondary Collision	5.492	0.0191	0.73
Barrier Type	1.250	0.2636
Restraint Use	5.222	0.0223
Vehicle Type	3.899	0.2726

Table 8 shows the odds ratios for secondary collision presence, vehicle type, barrier type, and occupant restraint usage for the developed binary logistic regression model. For the dichotomous secondary collision presence, barrier type, and restraint usage variables, the odds ratio shown was with respect to the other possible value (i.e. no secondary collision, metal barriers, and lap and shoulder belt usage). The vehicle type comparison group was passenger cars. The 95 percent confidence bounds on each odds ratio are also shown.

Table 8.

Odds Ratios and 95% Confidence Intervals for Injury Severity Model

Parameter	Value	Odds Ratio	95% CI
Secondary Collision	Yes	3.53	1.2 – 10
Vehicle Type	SUV	0.52	0.2 – 1.5
	Pickup Truck	0.42	0.2 – 1.1
	Van	0.74	0.1 – 6.1
Barrier Type	Concrete	1.96	0.6 – 6.4
Restraint Usage	Unbelted	3.64	1.2 – 11

The risk of serious occupant injury was found to increase by more than a factor of 3 with the presence of a secondary collision; this finding was statistically significant. A similar, statistically significant increase in occupant injury risk was found for unbelted occupants. Concrete barrier impacts were also found to increase serious occupant injury compared to metal beam barrier impacts, although this effect was not found to be statistically significant. Occupants of pickups, SUVs and vans were all found to have a decreased serious occupant injury risk compared to occupants of passenger cars. This vehicle type effect, however, was not statistically significant.

DISCUSSION

The results of the contingency table analysis suggest no difference in secondary collision rates for different barrier types. For vehicle type, SUVs were found to have the highest risk of involvement in a secondary collision following an initial traffic barrier impact. Approximately half of SUVs in the available data were involved in a secondary collision, compared to approximately 30 percent of passenger vehicles. For the entire data set, secondary collisions were found to occur in 34 percent (95% CI = 28.2 – 39.4) of tow-away level crashes where a barrier is the first object struck. This value is relatively consistent with previous literature. Data presented by Ray et al. (1987b) from New York and North Carolina suggested a secondary collision rate between 15 and 42 percent.

An analysis of the objects struck following an initial barrier impact revealed that more than 80 percent were an impact to another vehicle, an impact to a subsequent barrier, or vehicle rollover. The frequency of secondary impacts with trees and poles was substantially lower at 5 percent for all types of trees and poles combined. Examination of the distribution of the vehicle trajectory subsequent to a barrier impact revealed that there is considerable risk of a secondary impact for vehicles not ultimately involved in a secondary impact. Approximately half of the vehicles not involved in a secondary collision were found to have some level of intrusion into or across adjacent travel lanes. This finding was consistent to some degree with Ray et al. (1987b), however, the study by Ray et al. indicated a much higher percentage of vehicles crossing all lanes; 38 to 59 percent compared to approximately 10 percent for the current study.

The results of the binary logistic regression predicting secondary collision occurrence suggest that barrier penetration and vehicle type are the most important of the factors considered. SUVs were found to be more than twice as likely to be involved in a secondary collision compared to passenger cars.

This may have some implications on crash testing of traffic barriers which has traditionally assumed that the pickup truck vehicle to represent a more critical impact to a barrier. The results of this study suggest that this might not be the case, at least not for post-impact vehicle trajectory and vehicle stability, i.e. rollover. There was some evidence of increased occurrence of secondary collisions with concrete barrier impacts, length of need impacts, as well as impacts involving pickup trucks and vans. None of these secondary findings, however, were statistically significant.

The results of the injury prediction model suggest that the presence of a secondary collision significantly increases the risk of serious occupant injury. Compared to a single barrier impact, the occurrence of a secondary collision increases the risk of serious occupant injury by a factor of 3.5 (OR = 1.2 – 10). A very similar difference in risk was found between belted and unbelted occupants. Although not statistically significant, an interesting reversal was found with respect to vehicle type. Although light trucks such as SUVs, pickups and vans appear to be more likely to be involved in a secondary collision, the risk of occupant injury appears to be lower in these vehicles, compared to that of car occupants, for crashes involving barriers.

CONCLUSION

This study has investigated secondary collisions following an initial barrier impact in tow-away level crashes. Secondary collisions were found to occur in 34 percent (95% CI = 28.2 – 39.4) of tow-away level crashes where a traffic barrier was the first object struck. Secondary crashes were found to primarily involve an impact to another vehicle, an impact to another barrier, or a rollover; tree and pole impacts were found to represent a much smaller proportion of secondary impacts. Through a detailed analysis of vehicle trajectory, this study supports previous research that has suggested that risk of a secondary collision is substantial even for vehicles not ultimately involved in a secondary collision. Compared to a single barrier impact, the occurrence of a secondary collision was found to increase the risk of serious occupant injury by a factor of 3.5, equivalent to the risk difference found between a belted and unbelted occupant.