Analysis of the Effect of Speed Limit Increases on Accident-Injury Severities

Malyshkina and Mannering 1 Paper 08-0056 Analysis of the Effect of S peed Limit Incr eases on Accident-Injury Severities by Nataliya V. Malyshkina Research Assistant, Scho ol of Civil Engineering 550 Stadium Mall Drive Purdue University West Lafayette, IN 47907-2051 nmalyshk@purdue.edu Fred Mannering Professor, School of Civil Engineering 550 Stadium Mall Drive Purdue University West Lafayette, IN 47907-2051 flm@purdue.edu Number of words 4748 Tables and figures (2 x 250) = 500 Total = 5248 Prepared for the Transportation Research Board June 2007 Revised October 22, 2007 Malyshkina and Mannering 2 Abstract The influence of speed limits on roadway sa fety has been a subject of continuous debate in the State of Indiana and nationwid e. In Indiana, highway-related accidents result in about 900 fatalities and fo rty thousand injuries annually and place an incredible social and economic burden on the state. Still, speed limits posted on highways and other roads are routinely exceeded as individual drivers try to balance safety, mobility (speed), and the risks and penalties associated with law enfo rcem ent efforts. The speed-lim it/safety issue has been a matter of considerab le concern in In diana since th e state raised its speed limits on rural interstates and selected multilane highways on July 1, 2005. In this paper, the influence of the posted speed limit on the sever ity of vehicl e accidents is studied using Indiana accident data from 2004 (t he year before speed lim its were raised) and 2006 (the year after speed limits were raised o n rura l interstates and some multi-lane non-interstate routes). Statistical m odels of the injury s eve rity of different types of accidents on various roadway classes were estimated. The results of the model estimations sh owed that, for the speed limit ranges currently used, spe ed limits di d not have a statistica lly significan t effect on the severity of accidents on interstate hi ghways. However, for some non-interstate highways, higher speed limits were found to be associated with higher accident severities – suggesting that future speed limit changes, on non-interstate highways in particular, need to be carefully assessed on a case-by-case basis. Malyshkina and Mannering 3 Introduction The speed-limit changes that took eff ect on July 1, 2005, m ade Indiana the 30 th U.S. state to raise interstate speed limits up to 70 mi/h (raising them from 65 mi/h on rural interstates). Speed lim its were also increa sed on som e multilane non -interstate high ways. This intensified a statewide debate o n the tradeoff between highway mobility (sp eed) and safety. This debate has raged throughout the US for more than three decades since the passage of the Emergency Highway Ener gy Conservation Act in 1974, which mandated the 55 mi/h national maxim um speed limit on interstate highways in the US. State and federal speed-limit policy changes have been fueled by various research findings and subsequent legislation, such as the National Highway System Designation Act of 1995 that gave states complete free dom to set interstate speed lim its. With regard to safety, most research e fforts have concluded that the 1974-mandated 55 mi/h interstate speed limit had saved lives ( 1 , 2 ). This has been confirmed by some studies that have looked at recent speed limit increases on interstates. As an example, Kockelman and Bottom ( 3 ) found that a speed lim it increase from 55 to 65 m i/h resulted in roughly a 3% increase in the acc ident rate and a 24% increas e in the proba bility of a fatality once an accident occurred. For speed limit increases from 65 to 75 m i/h, they found a 0.64% increase in the acci dent rate and in a lower 12 % increase in the probability of fatal injury once an accident occurred. The authors speculated that these lower percentage increases from 65 to 75 mi/h speed lim its (relative to the 55 to 65 mi/h speed- limit increases) m ay have been the result of driv ers' heightened awarenes s of risk at higher speeds or that roads assigned the higher 75 m i/h in their study’s sam ple may have been inherently safer. However, other studies have contended that legislation-enabled speed- limit increases have actually saved lives. As an example, Lave and Elias ( 4 ) argued that the increase from 55 mi/h to 65 m i/h saved lives because of shifts in law enforcem ent resources, the ability of higher-speed-lim it inte r states to attract riskier drivers away f rom inherently more dangerous non-interstate highways, and possible reductions in speed variances. Understanding the magnitude of the safety impacts of increasing speed lim its, or even the direction of safety impacts (whether safety is im proved or compromised), rem ains a contentious subject because research has not been able to convincingly unravel the impacts of speed limit changes from the c onfounding effects of tim e-varying changes in factors such as highway enforcement, vehicle m iles traveled, vehicle occupancy, seat belt usage, alcohol use and driving, vehicle fleet mix (proportions of passenger cars, minivans, pickup trucks, and sport utility ve hicles), vehicle safety featur es (increasing adoption of air bags, antilock brakes, other active safety syst ems), speed limits on other road classes and in other states, driver expectations, and dr iver adjustment and adaptation to risk. In this paper, the recent increas e in Indi ana speed lim its was studied by undertak ing a statistical analysis to assess the effect that speed limits have on roadway safety by considering the relationship between speed lim its and observed accident-injury severities. In assessing the impact of speed limits on accident-injury severities, the injury level sustained by the most critica lly injured individual in an accident was used to conduct appropriate statistical tests to determ ine whet her the posted speed lim it had any significant Malyshkina and Mannering 4 effect on these injury severities and whether the possible effect changed after th e speed limits were raised. Methodology Accident severity (the most severe injury sustained by any vehi cle occupant in the accident) has discrete outcomes ranging from property damage only, to injury, to fatality. Given that these severity data are ordered re sponses from less severe to more severe, an ordered probability model would seem to be a natural approach and one that has been successfully applied to accident seve rity analysis by m any researchers ( 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 ) . H o w e v e r , a n a l t e r n a t i v e t o a n o r d e r e d model is the unordered probability approach that includes multinomial, nested and mixed logit m odels (see 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 ). Relative to ordered probability models , traditional multinomial, nested and m ixed logit structures do not account for the orde ring of injury-severity data. However, multinomial, nested and mixed logit m odels do offer a more flexibl e functional form. For example, the traditional m ultinomial logit can provide con sistent param eter estimates in the presence of the possible underreporting of accidents (for example, a known problem with police-reported data is the underreporti ng of property dam age only accidents). Also, they can relax the parameter restriction im posed by ordered probability models that does not allow a variable to sim ultaneously in crea se (or decrease) both high and low injury severities. This monotonic effect of variab les imposed by ordered probability m odels and its potential adverse consequences is discussed in Eluru and Bhat ( 22 ), Bhat and Pulugurta ( 23 ), and Washington et. al. ( 24 ). In this paper a multinomial logit model f or accident injury-severity outcomes was used. For the multinomial logit form ulation, a linea r function of covariat es that determine the probability of the accident injury outcome (of the m ost severely injured person) being reported as property damage only (no injury ), injury or fatality is defined as: in i n in Z ε = + β X , ( 1 ) where Z in is a linear function determ ining the probability of a ccident severity outcom e i for accident n, X n is a vector of measurable characteristics f or accident n that determine outcome i (such as speed limit, driver charac ter istics, roadway characteristics, environmental conditions, etc.), β i is a vector of estimable p arameters, and ε in is an error term that accounts for unobserved factors in fluencing resulting outcom es. McFadden ( 25 ) has shown that if ε in are assumed to be generalized extreme value distributed, the standard multinomial logit model results, () [ ] [] exp exp in n In I Pi = ∑ β X β X , (2) where P n ( i ) is the probability that accident n has accident-severity outcome i and I is the set of possible outcomes. This model is estim able by standard maximum likelihood methods ( 24 ). To assess the effect of the vector of estimated param eters ( β i ), elasticities are computed for each accident n ( n subscrip ting omitted) as Malyshkina and Mannering 5 () ( ) () ki Pi ki x ki Pi x E xP i ∂ =× ∂ , (3) where P ( i ) is the probability of discrete outcome i and x ki is the value of variable k for outcome i . This gives (using Equation 2 and 3), ( ) ( ) 1 ki Pi xk i k i E Pi x β ⎡⎤ =− ⎣⎦ , (4) where β ki is the estimated parameter associated with variab le x ki . Elasticity values can be roughly interpreted as the percent effect that a 1% change in x ki has on the accident- severity outcome probability P ( i ). The statistical analysis also used multiple statistica l tests to dete rmine if the accident data should be split by roadway type (rura l interstate, urban interstate, rural arterials, etc.) and accident type (single vehicl e, car colliding with car, car colliding with truck, etc). For this analysis a likelihood ratio test was used. The appropriate test statistic is ( 24 ), () () () 2 1 1 -2 - M m df M K m LL LL χ  = −× = ⎡⎤ ⎢⎥ ⎣⎦ ∑ ββ , (5) where LL ( β ) is the log-likelihood at converged values of β for the m odel estimated for th e whole data sample; LL ( β m ) is the log-likelihood of the mode l estimated for observations in the m th data subset (road/acciden t type combinations) and β m is the vector of parameters estimated for this m odel ( m = 1, 2, 3, ..., M ); M is the number of the data subsets; K is the number of param eters estimated for each m odel (i.e. K is the number of estim ated parameters in vectors β and β m ); and there are K M × − ) 1 ( degrees of freedom. The null hypothesis for Equation 5 is that the β 's in the M subset models are the same. If this hypothesis can be rejected with high confidence, estimation of se parate models for the data subsets is warranted. Data The accident data used in this stu dy we re from the Indiana Electronic Vehicle Crash Record System (EVCRS) which wa s launched in 2004 and includes available information on all accid ents investigated by Indiana police. The inform ation on accidents included into the EVCRS can be divided in to three major c ategories: roadway and environmental data (including weather, roadway and traffic conditions, roadway geometrics, posted speed limits, etc.); vehicle data (including inform ation on all vehicles involved in an accident, type and model of each ve hicle, vehicle mode l year, etc.); and occupant data (including information such as age and gender on all people who were involved in the accident and their injury stat us). These data gave 127 variables for each accident. Data were considered from 204,382 acci dents that were rep orted in the Indiana State accident databases for 2004 (the year be fore the speed limits we re raised) and from 182,922 accidents that were repor ted in the Indiana State acci dent databases for 2006 (the year after the speed limits were raised). Malyshkina and Mannering 6 Of the 182,922 accidents in the 2006 database , 65% occurred on locally m aintained streets and roads (city and county), 16.7% on state routes, 10.9% on US routes, and 7.4% on urban and rural interstates. Of these, 52.9% were two- vehicle accident s involving only passenger vehicles (passenger car and light truck s which include sport utility vehicles, vans and pickup trucks), 3.9% were vehicle accidents involving a large truck and a passenger vehicle, 31.1% were single vehicle accidents, and 12.1% were other accident types (involving 3 or more vehicles). This compar es closely to the 2004 da tabase that showed 69.3% occurred on loca lly maintain ed streets an d roads (city and county), 14.8% on state routes, 9.7% on US routes, and 6.2% on urban and rural interstates. Of these 2004 accidents, 54.7% were two-vehicle accidents involving only passenger v ehicles (passenger car and light trucks which include sport util ity vehicles, vans and pickup trucks), 4.8% were vehicle accidents involving a large truck and a passenger vehicle, 28.6% were single vehicle accidents, and 11.9% were other acci dent types (involving 3 or m ore vehicles) Of the accidents in 2006, unsafe speed was identified as the primary cause in 5.78% of the accidents. This compares to 7.28% of the 2004 accidents that listed unsafe speed as the primary cau se of th e accident (before the increase in speed limits). In ter ms of injury severity levels in 2006, 79.03% were property damage only, 20.56% were injury and 0.41% were fatality (the numbers for 2004 were 78.53% were property dam age only, 21.06% were injury and 0.41% were fatality). With regard to the effect of speed lim its, in 2006 unsafe speed was listed as the primary cause of the accident in 11.4% of the accidents on ro ads with 65m i/h and 70mi/h speed limits, 7.7% on roads with speed limits of 55mi/h and 60mi/h , 6.6% on roads with speed limits from 35m i/h to 50 mi/h, and 4.6% on roads with speed lim its of 30 mi/h or less. This compares with 2004 data that showed that unsafe speed was listed as the primary cause of the accident in 19.4% of th e accidents on roads with 65 mi/h speed limits (the maximum speed in 2004), 10.6% on roads w ith speed lim its of 55 mi/h and 60 mi/h, 7.5% on roads with speed limits from 35m i/h to 50 mi/h, and 6.0% on roads with speed limits of 30 mi/h or less. The co rrespond ing accid ent severity levels, by speed-limit category, for 2004 and 2006 are presented in Tabl e 1. This table shows some variation but generally similar numbers between the year s. However, to properly unravel this relationship, a m ultivariate analysis is needed to contro l for all of the m any factors that can potentially affect this relationsh ip. Estimation results Multinomial logit estimation results using the 2006 data are presented in Table 2 for the accident injury-severity models (the most severely injured occupant). The results for the 2004 data are similar and are not presented here to save space (pleas e see Malyshkina et. al ( 26 ) for a com plete presentation of th e 2004 results). Again, the possible outcomes are property damage only, injury and fatality. For estim ation purposes, the function determining property damage only (giv en by Equation 1) is set to zero without loss of generality ( 24 ). Based on the results of likelihood ratio tests, 34 different injury- severity models were estimated based on com binations of roadway type, roadway location and the number and types of vehicles involved in the accident. In addition to the speed- limit parameter estim ates shown in Table 2, the m odels included a wide variety of variables that were fou nd to significantly influ ence injury severity in cluding seasonal indicators, day-of-week indicat ors, peak-hour indicators, ot her time of day indicators, Malyshkina and Mannering 7 construction zone indicators, lighting conditions , precipitation indictor s (snow, rain, clear, etc.), pavement condition indictors (dry, wet, ice, etc.), m edian type, presence of vertical and horizontal curves, vehicle age, n umber of vehicle occupants, tra ffic control ind icators, driver age, driver gender. Turning specifically to the speed-limit para m eter estimates shown in Table 2, it is found that speed limits did not significantly aff ect accident-injury seve rities on interstate highways (this was also true of the model estimations base d on 2004 data). While the reasons for this are not know for certain, there could be a number of contributing factors to this finding. One is the issue of speed lim it compliance on inters tate highways. In a survey of Indiana drivers c onducted in the Fall of 2005 (a few months after Indiana interstate speed limits were raised), it was found that under free flow conditions drivers reported driving an average of nearly 11 mi/h over a 55 m i/h interstate speed limit, about 9 mi/h over a 65 mi/h inte rstate speed lim it and less than 8 mi/h over a 70 m i/h speed limit ( 27 ). Thus it appears that driver behavior is compressing the effect of speed limits (a 15 mi/h increase in speed lim its results in a less than 15 mi/h increas e in speeds). Also, this same survey found that the standard deviati on of self-reported free-flow speeds declined from roughly 6 mi/h on interstates posted 55 m i/h to about 5 m i/h on interstates posted 65 mi/h or 70 mi/h. The reduced stan dard deviation of speed may be m itigating the effect of the higher overall speeds on s everity. There co uld also be behavioral elem ents involved such as drivers becoming m ore alert (perhaps with lower reaction times) at highe r speeds and enabling them to take actions to reduce accident severity. Finally, the high design standards of the interstate system may be mitig ating the effects of higher speeds. All of the above factors seem to be sufficient to offset the physics involved with higher travel speed. Whether this would be true for speed limits ab ove 70 m i/h is an open question that our data can not support. In contrast to interstates, Table 2 s hows that for m any other highway types, increases in speed limits signi ficantly increase the likelihood th at the accident will result in an injury or fatality (this wa s also true of the model estima tions based on 2004 data). And, the elasticities show that th e effect can be reasonably large. For example, as sho wn in Table 2, for rural-county-road accidents involving a car or light truck with a heavy truck, a 1% increase in the speed lim its results in a 2.77% increase in the probability of f atality and a 2.35% increase in the probability of injury. Fo r rural-state-route acc idents involving a car or light truck with another car or light truck, a 1% increase in the speed limit results in a 11.9% increase in the probability of fatality and a 1.32% incre ase in the probability of injury. The accident-injury severity findi ngs on non-interstate hi ghways suggest that extreme caution needs to be exercised when rais ing th e speed limits on these roads. I t is speculated that the lack of access control, lo wer design standards and the greater demand placed on driver atten tion can make these highway s quite sensitive to spe ed-limit chan ges. Temporal Stability: Interstates The fact that speed limits on interstates were not found to affect the severity of accidents is worth a closer look. The previous analysis is based on cross-sectional data comparing the effect of different speed lim its on a class of roadways for a single year (2004 or 2006). This is an important consider ation because it controls for possible changes in vehicle safety features, enforcement levels and driver behavior th at may vary from one year to the next. Still, it may also be of interest to compar e 2004 and 2006 interstate Malyshkina and Mannering 8 accident-severity models to see if th e 2005 speed -limit in crease (and the other factors that may have varied over this time period) s ignificantly changed the estimated param eters in the accident severity models. To address this possibility, the followi ng are consid ered: 1) interstates that had 55 mi/h speed limits in 2004 and remained at 55 m i/h in 2006 (which were interstates in urban areas); and 2) inte rstates that were 65 mi/h in 2004 and increased to 70 mi/h in 2006 (which were those in rural areas). Again, applyi ng the likelihood ratio test, the appropriate statistical test is ( 24 ), () ( ) ( ) () 2004 2006 2 2004 2006 -2 all all df K K K LL LL LL χ ββ β  =+ − ⎡⎤ −− ⎣⎦ , (6) where LL ( β all ) is the log-likelihood at converged values of β for the model estim ated 2004 and 2006 data for the accident type being considered; LL ( β 2004 ) is the log-likelihood of the model estimated for 2004 observations; LL ( β 2006 ) is the log-likelihood of the model estimated for 2006 observations; K all is the number of param eters in β all ; K 2004 is the number of parameters in β 2004 ; and K 2006 is the number of parameters in β 2006 . The null hypothesis for Equation 6 is that the β 's in the 2004 and 2006 are the same. If this hypothesis can be rejected, it would indicate that the parameters have shifted from 2004 to 2006, suggesting that the increased speed lim it and possibly other influences (such as changes in driver behavior, changes in enfor cement levels, and changes in vehicle safety features) may have changed the effect of f actors that determine inju ry severity. For interstates that had the same 55 mi/h speed lim its in 2004 and 2006, the application of this likelihood ra tio test to the various m odels indicated that the accident severity parameter estim ates (for single and multivehicle crashes) did not signif icantly change from 2004 to 2006. In no case could th e null hypothesis be reject ed even at a very modest 70 percent confidence leve l. This indicates that th e "halo effect" of the increased speed limits (from 65 mi/h to 70 mi/h in rura l areas) did not significantly affect those urban interstates that rem ained at 55 mi/h. For interstates that were 65 mi/h in 2004 and increased to 70 mi/h in 2006, temporal stability tests again showed that acci dent severity param eter estimates (for single and multivehicle crashes) did not signifi can tly change from 2004 to 2006 (the null hypothesis could not be rejected in any of the cases even at the very modest 70 percent confidence level). The temporal stability of these intersta te models adds further evidence to support the cross-sectional finding that th e higher range of speed limits in effect on Indiana interstates in 2006 has not signifi cantly affected the severity of accidents. Summary and Conclusions The findings of this study are drawn from multinomial m odels of accident severity defined by the injury level of the most se verely injured person in the accident. The estimation results foun d that speed lim its did not significantly affect acciden t injury severities on interstate highways. This is an important finding because the July 1 , 2005 increase in maxim um interstate speeds from 65 mi/h to 70 mi/h has been the focus of considerable media atten tion. One can speculate that th is finding is a res ult of a number of factors including possible reductions in speed variance as speed limits increase, driver responses to higher speed limits and the high design standards of the interstate system which appear to be able to accommodat e modest increases in speed lim its. Malyshkina and Mannering 9 For non-interstate highways, the results ar e quite different. For non-interstate highways, the accident data show that higher speed lim its ar e associated with a greater likelihood of injury and/or fatality on some (but not all) roadway types (county, state, city and US routes) and accident t ypes (single and two-vehicle). This study's findings have a number of imp lications for speed lim it polices in the State of Indiana. With regard to interstate speeds, the findings suggest that the effect of speed limits on accident injury sev erity are not necessarily a cause for concern for the speed-limit ranges that were cons idered in this study (55 mi/h to 70 m i/h). Whether this finding would hold true if speed limits were in creased further to 75 m i/h or 80 mi/h (both of which would exceed the 70 mi/h inters tate-standard design sp eed ) remains an open question. To be sure, the additional speed would increase stopping distances and the energy that would need to be dissipated in the accident. Also, at some point, higher speed limits may star t increasing the varia nce in driver speeds as som e drivers continue to drive at or above the speed limit while others driv e belo w the speed limit because the speed lim it may have been raised above th eir "optimum " speed. With these factors considered (along with other factors that may come into play such as variations in driver behavior in response to speed limits), there is likely a poin t beyond which higher s peed limits will significantly increase the severity of acciden ts on interstates. With regard to speed limit policies on road ways other than inte rstate highways, our results suggest that considerab le caution should be exercised. The findings show that, on some non-interstate roadway and accident type com bi nations, higher speed lim its significantly increase th e likelihood that accident s will result in injuries and fatalities. Thus, changing speed limits on non-interstate highways should be done on a case-by-case basis taking into account past accident history as well as th e specific geometrics and access control of the facility, as these factors can vary widely even with in the same class of highways (non-interstate). Acknowledgments The authors were supported by the Indian a Department of Transportation/Joint Transportation Research Program (JTRP) Projec t – SPR 3030. The contents of this paper reflect the views of the authors, who are res ponsible for the facts a nd th e accuracy of the data presented herein. The contents do not n ecessary reflect the official views or policies of the Federal Highway Administration and the Indiana Department of Transportation, nor do the contents constitute a standard, specif ication, or regulation. The comments and suggestions of Brad Steckler are gratefully acknowledged. Malyshkina and Mannering 10 References 1. National Highway Traffic Safety Administ ration. The Effects of the 65 mph Speed Limit Through 1990: A Report to Congress. US Department of Transportation, Washington, DC, 1992. 2. Farmer, C., R. Retting, and A. Lund. Cha nges in Motor Vehicle Occupant Fatalities After Repeal of the Nati onal Maximum Speed Lim it. Accident Analysis and Prevention , Vol. 31, No. 5, 1999, pp. 537-543. 3. Kockelman, K., and J. Bottom. Safety Im pact s and Other Implications of Raised Speed Limits on High-Speed Roads. National C ooperative Research Program , Project 17-23, Washington, DC, 2006. 4. Lave, C., and P. Elias. Did the 65 mph Speed Lim it Save Lives? Accident Analysis and Prevention , Vol. 26, No. 1, 1994, pp. 49-62. 5. O’Donnell, C. and D. Connor. Predicting th e Severity of Motor Vehicle Accident Injuries Using Models of Ordered Multip le Choice. Accident Analysis and Prevention , Vol. 28, No. 6, 1996, pp. 739- 753. 6. Duncan C., A. Khattak, and F. Council. A pplying the Ordered Pr obit Model to Injury Severity in Truck-Passenge r Car Rear-End Collisions. Transportation Research Record 1635, 1998, pp. 63-71. 7. Renski, H., A. Khattak, and F. Council. Impact of Speed Limit Increases on Crash Injury Severity: Analysis of Single-Vehi cle Crashes on North Carolina Interstate Highways. Transportation Research Record 1665, 1999, pp. 100-108. 8. Khattak, A. 1. Injury Severity in Multi-Vehic le Rear-End Crashes. Transportation Research Record 1746, 2001, pp. 59-68. 9. Kockelman, K., and Y.-J Kweon. Driver Inju ry Severity: An Application of Ordered Probit Models. Accident Analysis and Prevention , Vol. 34, No. 4, 2002, pp. 313-321. 10. Khattak, A., D. Pawlovich, R. Souleyrette, and S. Hallmark. Factors Related to More Severe Older Driver Traffic Crash Injuries. Journal of Transporta tion Engineering , Vol. 128, No. 3, 2002, pp. 243-249. 11. Kweon, Y.-J., and K. Kockelman. Overall Inju ry Risk to Different Drive rs: Combining Exposure, Frequency, and Severity Models. Accident Analysis and Prevention , Vol. 35, No. 3, 2003, pp. 414-450. 12. Abdel-Aty, M.. Analysis of Driver Injury Se verity Levels at Multiple Location s Using Ordered Probit Models. Journal of Safety Research , Vol. 34, No. 5, 2003, pp. 597-603. 13. Shankar, V., and F. Mannering. An Explor atory Multinomial Logit Analysis of Single- Vehicle Motorcycle Accident Severity. Journal of Safety Research , V ol. 27, No. 3, 1996, pp.183-194. 14. Shankar, V., F. Mannering, and W. Barfield. Statistical Analysis of Accident Severity on Rural Freeways. Accident Analysis and Prevention , Vol. 28, No. 3, 1996, pp. 391- 401. 15. Chang, L.-Y., and F. Mannering. Analysis of Injury Severity and Vehicle Occupancy in Truck- and Non-Truc k-Involved Accidents. Accident Analysis and Prevention , Vol. 31, No. 5, 1999, pp. 579-592. 16. Carson, J., and F. Mannering. The Eff ect of Ice Warning Signs on Accident Frequencies and Severities. Accident Analysis and Prevention , Vol. 33, No. 1, 2001, pp. 99-109. Malyshkina and Mannering 11 17. Lee, J., and F. Mannering. Impact of Roadsi de Features on the Frequency and Severity of Run-Off-Roadway Accident s: An Empirical Analys is. Accident Analysis and Prevention Vol. 34, No. 2, 2002, pp. 149-161. 18. Ulfarsson, G., and F. Mannering. Differences in Male and Fem ale Injury Severities in Sport-Utility Vehicle, Minivan, Pickup and Passenger Car Accidents. Accident Analysis and Prevention , Vol. 36, No. 2, 2004, pp. 135-147. 19. Khorashadi, A., D. Niemeier, V. Shankar, and F. Mannering. Di ffe rences in Rural and Urban Driver-Injury Severities in Accident s Involving Large Tru cks: An Exploratory Analysis. Accident Analysis and Prevention , Vol. 37, No. 5, 2005, pp. 910-921. 20. Savolainen, P., and F. Mannering. Probabi listic Models of Motorcyclists' Injury Severities in Single- and Multi-Vehicle Crashes. Accident Analysis and Prevention , Vol. 39, No. 5, 2007, pp. 955-963. 21. Milton, J., V. Shankar, and F. Mannering. Highway Accident Seve rities and the M ixed Logit Model: An Exploratory Empirical Analysis. Accident Analysis a nd Prevention , Vol. 40, No. 1, 2008, pp. 260-266. 22. Eluru, N., and C. Bhat. A Joint Econometri c Analysis of Seat Belt Use and Crash- Related Injury Severity. Accident Analysis and Prevention , Vol. 39, No. 5, 2007, pp. 1037-1049. 23. Bhat, C., a nd V. Pulugurta. A Comparis on of Two Alternate Behavioral Choice Mechanisms for Household Auto ownership Decisions. Transportation Research P art B , Vol. 32, No. 1, 1998, pp. 61-75. 24. Washington, S., M. Karlaftis, and F. Manne ring. Statistical a nd Econometric Methods for Transportation Data Analysis. Chap man and Hall/CRC, Boca Raton, FL, 2003. 25. McFadden, D. Econometric Models of Probabilistic Choice, In A Structural Analysis of Discrete Data with Econometric Applica tions , Manski and D. McFadden, editors, The MIT Press, Cambridge, MA, 1981. 26. Malyshkina, N., F. Mannering, and S. La bi. Influence of Speed Limits on Roadway Safety in Indiana. Prepared for the Join t Transportation R e search Program , Indiana Department of Transportation, Federal Highway Adm inistration, Project No. C-36- 59VV, 2007. 27. Mannering, F. Speed Limits and Safety: A St atistical Analysis of Driver Perceptions. Transportation Research Board 85 th Annual Meeting, Compe ndium of Papers CD- ROM, Washington, DC, 2007. Malyshkina and Mannering 12 List of Tables Table 1. Indiana accident injury-severity d i stributions by posted speed limits in 2004 and 2006. Table 2. Speed limit param e ter and elasticity estimates (com puted for statistically significant param eters) for accident severity models based on 2006 Indiana accident d ata. Malyshkina and Mannering 13 Table 1. Indiana accident injury-severity d i stributions by posted speed limits in 2004 and 2006. Injury Severity Level Speed limit Property Damage Only Injury Fatality 2004 posted 65 mi/h 81.7% 17.7% 0.6% 2006 posted 65 mi/h or 70 mi/h 81.9% 17.4% 0.7% 2004 posted 55 mi/h and 60 mi/h 76.7% 22.3% 1.1% 2006 posted 55 mi/h and 60 mi/h 77.8% 21.3% 0.9% 2004 posted 35 mi/h to 50 mi/h 74.5% 25.2% 0.4% 2006 posted 35 mi/h to 50 mi/h 75.3% 24.3% 0.4% 2004 posted 30 mi/h or less 80.6% 19.2% 0.2% 2006 posted 30 mi/h or less 82.1% 17.8% 0.2% Malyshkina and Mannering 14 Table 2. Speed limit param e ter and elasticity estimates (com puted for statistically significant param eters) for accident severity models based on 2006 Indiana accident d ata. Speed limit parameter estimate ( t -ratio) Model (C=cars, LT=Light trucks; HT=Heavy Trucks) fatality injury Fatality Elasticity Injury Elasticity (C/LT)+(C/LT) 0.0396(5.48) 0.0396(5.48) 1.61 1.20 (C/LT)+(HT) 0.0648(3.06) 0.0648(3.06) 2.77 2.35 Rural one vehicle 0.00506(2.04) 0.00506(2.04) 0.24 0.19 (C)+(C) 0.00689(0.00) 0.00507(.321) (C)+(LT) 0.0231(0.00) 0.0613(2.43) 1.80 (LT)+(LT) .0110(0.00) 0.0269(0.84) (C/LT)+(HT) -0.5454(-.518) -0.0288(-0.30) County road Urban one vehicle -0.0852(-1.23) 0.000725(0.07) (C/LT)+(C/LT) 0.103(1.28) 0.00872(0.88) (C/LT)+(HT) 0.150(0.91) 0.00133(0.06) Rural one vehicle -0.0237(-1.55) -0.0237(-1.55) (C/LT)+(C/LT) 11.04(0.00) -0.00108(-0.14) (C/LT)+(HT) -0.00188(-0.01) 0.0120(0.52) Interstate Urban one vehicle 0.00776(0.20) 0.00384(0.48) (C/LT)+(C/LT) 0.248(3.48) 0.0416(3.25) 11.9 1.32 (C/LT)+(HT) 0.127(2.50) 0.127(2.50) 5.79 5.36 Rural one vehicle 0.0636(2.34) 0.0127(2.25) 3.34 (C/LT)+(C/LT) 0.251(3.35) .0290(7.95) 9.40 0.84 (C/LT)+(HT) 5.60(0.00) 0.452(1.73) State route Urban one vehicle 0.0268(0.42) -0.0115(-0.83) (C/LT)+(C/LT) 0.0414(6.13) 0.0414(6.13) 1.46 1.12 (C/LT)+(HT) 0.0185(0.00) 0.540(1.43) Rural one vehicle -0.0800(-1.46) -0.00409(-0.38) (C)+(C) 0.0251(6.33) 0.0251(6.33) 0.81 0.63 (C)+(LT) 0.0218(4.65) 0.0218(4.65) 0.73 0.56 (LT)+(LT) 0.0343(4.20) 0.0343(4.20) 1.14 0.87 (C/LT)+(HT) 0.0284(2.34) 0.0284(2.34) 0.94 0.83 City street Urban one vehicle 0.00968(0.38) -0.00128(-0.24) (C/LT)+(C/LT) 0.0644(1.32) 0.0272(1.84) (C/LT)+(HT) 0.0608(3.07) 0.0608(3.07) 3.12 2.28 Rural one vehicle 0.0137(1.56) 0.0137(1.56) (C/LT)+(C/LT) 0.0154(2.14) 0.0154(2.14) 0.61 0.44 (C/LT)+(HT) 0.0586(3.60) 0.0586(3.60) 2.33 2.02 US route Urban one vehicle 0.0327(0.45) 0.0134(.878)