Determinants of Traffic Fatalities in Sweden

Transcription

1 Advances in Managemen & Applied Economics, vol. 7, no. 2, 2017, 1-21 ISSN: (prin version), (online) Scienpress Ld, 2017 Deerminans of Traffic Faaliies in Sweden Sherzod Yarmukhamedov 1 Absrac This paper presens an empirical invesigaion of he casualies of road raffic faaliies in Sweden from he perspecive of differen socio-economic facors. We use he annual daa on raffic faaliies, income level, vehicle park, populaion, raffic offence, fuel price and unemploymen rae over he period Paper addresses he mehodological issues relaed o he nonsaionariy problem and reflecs he oucome in he model specificaion. Resuls indicae ha driving license revocaions due o he drunk driving are posiively associaed wih raffic faaliies, which implies ha police srenghens he enforcemen of raffic laws afer he increase in raffic faaliy rae. Moreover, an increase in unemploymen rae reduces raffic faaliies hrough a negaive impac on vehicle ownership. A las, a raffic faaliy rae ends o decrease over ime due o he advances in vehicle design, highway qualiy, medical care and oher raffic safey measures. JEL classificaion numbers: C22, C51, O18, R41 Keywords: raffic faaliy, acciden, uni roo, ime series 1 Inroducion The problem of deah and injury due o he road raffic accidens yield medical, social and economic problems o sociey which requires advancing insiuional responsibiliy for road safey. Therefore, providing a greaer safey on roads and exenuaing he harmful effecs of vehicles on human healh is one of he global policy issues o be prioriized. The World Healh Organizaion (WHO) repors abou 1.3 million deahs in road accidens annually, which accouns for 2.2% of oal deahs in he world. The 1 The Swedish Naional Road and Transpor Research Insiue (VTI), Sweden. Aricle Info: Received: December 2, Revised: December 30, Published online: March 1, 2017

2 2 Sherzod Yarmukhamedov projeced rends in global moraliy shows ha here will be a 28% increase in global deah due o he injury beween 2004 and 2030, which mainly resuls from he increasing number of road acciden deahs. Road raffic accidens deahs are projeced o increase o 2.4 million in 2030, primarily due o he increased moor vehicle ownership and use associaed wih economic growh in low and middle income counries. This implies ha, road raffic accidens will raise from ninh leading cause of deah globally in 2004 o he fifh in The economic consequences of road raffic accidens are also significan, in erms of boh los produciviy and all healhcare coss as an immediae corollary. The World Bank saes ha road accidens cos a counry approximaely 1 o 3% of is annual Gross Naional Produc (GNP). These expenses consiue a significan share of naional economy, which may impede he economic and social developmen of he counry. The global cos is esimaed o be USD 518 billion per year. Low-income and middle-income counries accoun for USD 65 billion, more han hey receive in developmen assisance. By conras, a modes amoun of resources is invesed in prevening he road accidens and injuries. The esimaed global research and developmen funding for road raffic accidens comprise USD millions. Comparaively lile is spen on implemenaion, even hough many inervenions ha would preven he accidens and injuries are well known, well esed, cos-effecive and publicly accepable (WHO, 2004). Idenifying he facors ha can shed ligh on he incidence of raffic accidens and a horough analysis of he circumsances ha lead o raffic accidens is fundamenal in developing raffic safey policies. Therefore, invesigaing he magniude, scope, characerisics and consequences of road raffic accidens can increase he effeciveness of inervenions. In he presence of complexiy of he road raffic sysem, analyzing he facors ha has an impac on road raffic faaliy is purposed in his research. Therefore, we inegrae he findings of pas works wih quaniaive analysis made on differen counries o empirically analyze he raffic accidens in Sweden. We model a raffic faaliy rae as a funcion of several socio-economic facors and use he aggregae annual daa for he period Resuls indicae ha driving license revocaions due o he drunk driving migh be posiively associaed wih raffic faaliy rae. This unexpeced posiive relaionship migh be explained by more sringen enforcemen of raffic safey regulaions by police afer raffic accidens. We also find ha as unemploymen rae increases, raffic faaliies end o decline due o worsening of he purchasing power, which may affec he affordabiliy of vehicle ownership. Consequenly, he fewer vehicles in raffic may decrease he probabiliy of raffic accidens. Finally, we find ha he raffic faaliy is likely o decrease over ime due o he improvemen of driving skills, vehicle design, highway qualiy, medical care and oher raffic safey measures. The conribuion of his sudy migh be refleced in hree regards. Firs, we address some mehodological issues ha are presen in previous sudies. Second, he oucome of his sudy may faciliae designing he efficien policy inervenions o improve raffic safey by considering he facors ha have a significan impac on

3 Deerminans of Traffic Faaliies in Sweden 3 raffic faaliy. A las, his is he firs empirical sudy ha analyzes he incidence of raffic faaliies using socio-economic facors for he case of Sweden. The paper is consruced as follows. Secion 2 provides a lieraure review. The Swedish passenger car raffic performance is discussed in Secion 3. Model specificaion and daa descripion are presened in Secions 4 and 5, respecively. Secion 6 provides he resuls of a uni roo es, mulicollineariy es, muliple regressions, as well as he resul discussion. Secion 4 concludes he paper. 2 Lieraure review Traffic safey lieraure discusses several socio-economic facors ha migh be he deerminans of raffic faaliies. For insance, he level of economic performance of a counry may have an impac on raffic faaliy rae. According o he World Healh Organizaion (2004), while developing counries own only abou 32 % of he world s moor vehicles, hey accoun for 85 % of he 1 million individuals killed and 24 million injured in road raffic accidens each year (Anbarci e al., 2006). Bishai e al. (2006) invesigae he impac of he Gross Domesic Produc (GDP) per capia on raffic faaliy rae in he low-income and wealhy counries. The resuls indicae ha a 10% increase in he GDP in a low-income counry is expeced o raise he number of accidens by 7.9%, he number of raffic injuries by 4.7%, and he number of faaliies by 3.1% hrough a mechanism ha is independen of a populaion size, vehicle couns, oil use, and roadway availabiliy. Increases in GDP in rich counries appear o reduce he number of raffic faaliies, bu do no reduce he number of accidens or injuries, all else equal. Greaer perol use and alcohol use are relaed o more raffic faaliies in rich counries. In low-income counries, a rise in raffic accidens, injuries, and faaliies accompanies economic growh. Afer a cerain hreshold in economic growh per capia, here is no longer an increase in he number of faaliies, hough he number of accidens and injuries coninue o increase wih growh. Examining he impac of income growh on he raffic faaliy raes, Kopis and Cropper (2003) argue ha he relaionship beween moor vehicle faaliy rae and per capia income a firs increases wih per capia income, reaches a peak, and hen declines. The reason is ha a low income levels, he rae of increase in moor vehicle oupaces he decline in faaliies per moor vehicle. A higher income levels, he reverse occurs. Therefore, he observed paerns illusrae a decline in faaliies in wealhy counries, whereas an increase in faaliies in developing counries. Anbarci e al. (2006) explaining his phenomenon argue ha in a low-income counry wih low moorizaion raes, here is a low risk of raffic faaliy. However, as per capia income increases, he moorizaion raes also increase which may resul in an increase in raffic faaliy. Evenually, as per capia income passes a cerain hreshold, several facors come ino play which

4 4 Sherzod Yarmukhamedov may slow a growh in negaive exernaliy. This migh be explained by he fully-developed roadway infrasrucures, pedesrian walkways ha separae walkers from moor vehicles, safey inspecions of vehicles, sea bel requiremens, regulaions on alcohol consumpion while driving, and effecive enforcemens of raffic laws and regulaions. However, when i comes o he infrasrucure developmen, Noland (2003) argues ha he infrasrucure improvemens are no effecive in reducing faaliies and injuries. Insead, he demographic changes in age cohors, increased sea-bel use, reduced alcohol consumpion and increases in medical echnology may lead o he overall reducions in raffic faaliies. Law e al. (2011) perform an empirical analysis of he Kuznes curve relaionship beween he economic growh, measured by per capia income, and raffic faaliies across 60 counries. The resuls show he presence of Kuznes curve relaionship beween per capia income and raffic faaliies, and sudy concludes ha he medical and echnology improvemens has a significan effec in decreasing he number of raffic faaliies. The negaive consequences from he road accidens are regarded as socio-economic coss from he sociey s perspecive. The cos esimaes should hen reflec he social uiliy of decreasing he number of faal road raffic accidens. The coss per faal injury are usually defined as direc (coss for healh care, propery damages and adminisraion) and indirec coss (los producive capaciy) plus a value of safey per se (Trawen e al., 2002). One of he imporan causes of raffic faaliies is speeding. Blows e al. (2005) argue ha hose who has repored frequenly racing a moor vehicle for exciemen or driving a 20 km/h or more over he speed limi, and hose who has received he raffic convicions over he pas 12 monhs, have been beween wo and four imes more likely o be injured while driving over he same ime. Driving unlicensed has been a risk facor for older bu no younger drivers, and driving a 20 km/h or more above he speed limis has been a sronger risk facor for younger (<25 years) han older drivers. Tarko (2009) proposes a new model of driver-preferred speeds derived from he assumpion ha drivers rade-off a porion of heir safey for a ime gain. The densiy of inersecions, land developmen along he road, and he presence of sidewalks are he idenified prominen risk percepion facors. Moreover, he speed limi seems o encourage slow drivers o drive faser and fas drivers o drive slower. The fracion of drivers driving under he alcohol influence is anoher leading cause of raffic faaliies. Drivers wih alcohol in heir blood are seven imes more likely o cause a faal crash, while legally drunk drivers pose a risk hireen imes greaer han sober drivers (Levi and Porer, 2001). Sudying he effecs of sae alcohol policies on moor vehicle faaliies for children, Sen and Campbell (2010) find ha he number of faaliies among he child moor vehicle occupans is srongly correlaed wih he alcohol use measured a he sae level and ha he adminisraive license revocaion policies and he higher beer ax raes appear o consisenly reduce such faaliies. For wo of he hree age groups, he beer ax

5 Deerminans of Traffic Faaliies in Sweden 5 raes are appeared o reduce faaliies during he nigh raher han he day. However, a zero olerance and blood alcohol concenraion limi laws do no seem o have any saisically significan effecs on faaliies. Islam and Mannering (2006) sudy he severiy of acciden injury wih respec o differences in gender and age. Resuls show he increased likelihood of faaliy for he young and middle-aged male drivers when hey have passengers (which was no a significan facor for females); he increased likelihood of faaliy for he young and older male drivers when driving he vehicles less han 5 years old (which was no a significan facor for females); he increased likelihood of injury for he middle-aged female drivers when driving vehicles aged 6 years and over (which was no a significan facor for males); and he increase in faaliy likelihoods for he older males aged beyond 65 years (which was no a significan for females). Simoncic (2001) shows ha i is more likely ha a severe injury or faaliy can be he oucome of he raffic acciden if he vehicle owner is aged 25 years or less compared o he case where he driver is older. This implies ha he severiy of acciden consequences migh be dependen on he drivers age. Analyzing he older road users propensiy o have accidens, Mori and Mizohaa (1995) argue ha he reduced physical and menal funcions wih he advancing age give improper driving characerisics o he elderly drivers, bu are no deerminisic facors of quiing he driving. Some elderly drivers who do no recognize or do no wan o recognize he reducion in heir physical condiion may impose a risk of raffic acciden. Probably, driving a vehicle suppors heir posiive social aciviy and puing a resricion on heir social aciviies by revoking a driving license migh evoke he social concerns. The risk of faaliy is also differen depending on he mode of ravel. For insance, Beck e al. (2007) show ha faaliy raes are highes for he moorcycliss, pedesrians and cycliss, while he nonfaal injury raes are highes for he moorcycliss and cycliss. Elvik and Vaa (2004) compare he injury risks by mode of ravel in six European counries. They calculae he injury raes per kilomeer ravelled and found ha relaive o he car occupans, pedesrians, bicycliss, and moorcycliss are a increased risk and bus occupans are a decreased risk. Simoncic (2001) also demonsraes ha he pedesrians and moorcycliss are a higher risk han he cycliss. Wong e al. (2009) idenifying he groups ha are a he faaliy risk show ha hose aged over 60 years are as four imes as likely o be pedesrian faaliies. Conversely, he risk of faaliies involving pedesrians and cycliss is reduced for males. However, males are a increased risk of faaliies involving he moorcycliss, scooer and pillion riders, whereas such risk is reduced for hose aged 30 and over. Clark and Cushing (2004) analyze he relaionship beween he populaion densiy and raffic faaliy raes in urban and rural areas. The resuls show ha he sae populaion densiy is moderaely srong predicor of rural bu no urban raffic faaliies.

6 6 Sherzod Yarmukhamedov Hya e al. (2009) sudy he relaionship beween raffic faaliies and changes in gasoline prices. High gasoline prices decrease he number of raffic faaliies for auomobiles bu he faaliy rae increases for moorcycliss. This suggess ha he increase in he price of gasoline may induce car drivers o swich o he moorcycles as anoher mode of ransporaion. Grabowski and Morrisey (2004) show ha a 10-cen increase in gasoline prices increases he moor vehicle faaliies by 2.3% over a 2-year period. By he same oken, Wilson e al. (2009) argue ha rising he gasoline prices may resul in over 1500 addiional moorcycle faaliies annually for each dollar increase in he gasoline prices. Wenzel and Ross (2005) sudy he influence of a vehicle ype and vehicle model on raffic faaliies. Auhors find ha mos car models are as safe o heir drivers as mos spor uiliy vehicles (SUV). However, he SUV and pickup rucks impose higher faaliy risk on he drivers of oher vehicle ypes, and his faaliy risk increases wih he vehicle size. Lam (2003) invesigaes he risk paerns of he drivers wih differen license saus. The resuls indicae ha females are a higher risk of being died or injured in a raffic acciden han males. If no conrolling for he driving experience and ages, he main cause for accidens are drivers own behaviors and driving condiions. Special road feaures increase he risk of an acciden injury for drivers wih a learner and provisional license saus. Alcohol has an impac on he fully licensed drivers aged years. 3 Passenger car raffic performance in Sweden A high level of economic developmen and qualiy of life conribues o he increase in moorizaion rae in Sweden. Beween 1992 and 2009, he rae of moorizaion has increased from 487 vehicles 2 o 565 vehicles per 1000 inhabians (Fig. 3.1). The rae of moorizaion has decreased by 3.30% during , and sharply increased by 15.15% during In general, he rae of moorizaion is rising, and during he las 18 years he increase was 16.17%. The increasing rae of moorizaion leads o a higher level of exernaliies in he form of air polluion, noise, road congesion, raffic accidens ec. For insance, he acciden rae has increased from 179 o 200 per inhabians during he period from 1992 o 2009 (Fig. 3.2). As he figure shows, a relaively sable acciden rae is observed during , while a sudden increase by 15.4% occurs beween and remains sable in he subsequen years. Overall, he raffic acciden rae has on average increased by 11.58% during 1992 and Henceforh, he passenger cars are denoed as vehicles.

7 Deerminans of Traffic Faaliies in Sweden 7 The consequences of raffic accidens are illusraed by he rae of injuries and faaliies ha occurred on he road. Wih he sligh flucuaions, he number of injuries has been increasing unil 2002, resuling in 543 raffic injuries per individuals (Fig. 3.3). Thereafer, here is a drasic decrease by 41% beween 2003 and The number of raffic faaliies has been decreasing unil 1998, aferwards, here was an increase by 10.8 and 5.5% during and , respecively (Fig. 3.4). In general, he rae of faaliies is decreasing, and during 1992 and 2009 a decline in raffic faaliies became 56.12%. Noe ha despie he increase in he moorizaion and acciden raes, he raes of raffic injuries and faaliies are decreasing. In 1997, Sweden has adoped a policy of Vision Zero which became he foundaion for road safey operaions in he counry. The objecive of he policy is ha no one should be killed or seriously injured in he road raffic. Road raffic is a complex sysem consising of differen componens such as he roads, vehicles and road users, where a consrucive ineracion among hem is essenial o ensure he raffic safey. The obvious resul of he Vision Zero is he change in he road environmens in Sweden. The cenral median barriers and roundabous have become much more common, as have differen ypes of speed reducing measures in he buil-up areas. The resuls achieved so far indicae he evidence of a reducion in he number of raffic faaliies in he recen years, hough he raffic is

8 8 Sherzod Yarmukhamedov increasing (Vägverke, 2006). This can be explained by he realizaion of he road safey measures developed by he naional road safey research insiuions. For insance, Elvik e al. (2009) discuss he benefis of road safey research in Sweden. Auhors conclude ha he oucomes of he road safey research serve for he benefi of sociey, and hese benefis by a wide margin ouweigh he cos of he research. 4 Model To analyze he impac of various socio-economic facors on he raffic faaliy rae, we use a muliple regression model ha considers he income level, populaion, vehicle park size, unemploymen rae, raffic offence, fuel price and ime rend o explain he incidence of raffic faaliies. This implies ha he general relaionship is represened by he following equaion: ln( FATAL ) ln( INC ) ln( POP ) ln( VEHP ) ln( TROFF ) FUEL UNEMP TREND The dependen variable FATAL represens he annual number of deahs ha occurred on he road; INC- he annual income level; POP- he amoun of populaion; VEHP he size of he vehicle park; TROFF- he amoun of raffic offence; FUEL a fuel price; UNEMP- he rae of unemploymen; TREND- a ime rend variable. To faciliae he inerpreaion of he coefficien esimaes, he variables income, populaion, vehicle size and raffic offence are log ransformed. Before esimaing a raffic faaliy model presened above, we need o conduc wo diagnosic ess. The firs one is a uni roo es which is used o examine wheher he variables in he model are saionary or no. The esing procedure requires he idenificaion of he naure of nonsaionariy, which is horoughly discussed in he nex secion. The accuracy in conducing his es is imporan, since he consequences of using nonsaionary ime series are unreliable coefficien esimaes from he leas square esimaor, es saisics and predicions. Some sudies underprioriize he imporance of his es, which leads o he problem of spurious regression (Wooldridge, 2013; Hill e al., 2001). The second es is a mulicollineariy es where he srengh of a relaionship beween explanaory variables is examined. If correlaion beween wo variables is oo high (larger han 0.7), he individual impac of variables becomes difficul o idenify. In ha case, one of he highly correlaed variables should be dropped. Based on he resuls of a uni roo and mulicollineariy ess, we modify our raffic faaliy model.

9 Deerminans of Traffic Faaliies in Sweden 9 5 Daa The sample consiss of annual daa over he period from 1992 o The sudy is performed on he example of Sweden, and he vehicles considered are passenger cars, so ha he rucks, racors, busses, moorcycles ec. are excluded. The daa are consruced from raffic faaliy rae, income level, populaion, vehicle park size, unemploymen rae, raffic offence and fuel price. The sources of he daa are he Saisics Sweden, he Swedish Insurance Federaion, he Transpor Analysis, Bil Sweden, he Swedish Transpor Agency, he Swedish Peroleum Insiue, and he Swedish Insiue of Public Healh. A descripion of he explanaory variables and heir expeced impac on a raffic faaliy rae is discussed below. The income level of a counry is refleced by is GDP per capia, he Swedish crowns (SEK), curren prices. An increase in income level would increase he demand for safey and consumpion aciviy. A rise in income level increases he probabiliy of an acciden due o accessibiliy of using a vehicle, bu he probabiliy of faaliy from an acciden would decline because of he increase in demand for safey (Pelzman, 1975; Zlaoper, 1989, 1991). Therefore, i is expeced ha he income level variable (INC) is posiive. The amoun of populaion wih driving licenses migh have a direc influence on raffic faaliy. The proporion of young drivers possessing a driving license beween he ages of 18 and 24 has he highes acciden raes han oher age groups. The naure of he group specific acciden raes shows ha acciden raes are highes for he younges drivers, decline wih age, hen rise for he older drivers. This can be explained by he younger drivers greaer risk-aking behavior and elderly drivers declining physical capabiliies (Anbarci e al., 2006). Therefore, we include wo variables o accoun for he young driver groups aged (POP1), and elderly driver group aged 66 and over (POP2). To noe, compared wih many previous sudies ha use he general age-classified populaion daa, we have an access o more deailed daa on he number of issued driving licenses caegorized by age, so ha i would be possible o encompass he group ha poenially causes raffic accidens. Furhermore, we admi ha drivers may have differen caegories of driving licenses (A, B, C ec.). However, as curren sudy focuses on he passenger cars only, we have proxied hese variables by he number of drivers ha has a B-caegory driving license. I is expeced ha he raffic faaliy raes are posiively associaed wih hese wo age groups. According o he Swedish legislaion, every regisered vehicle in Sweden can have he saus of being eiher an acive-license or inacive-license vehicle. To use a vehicle in raffic, i mus have a saus of acive-license vehicle, which implies ha driver mus pay for vehicle insurance and ax. Before cancelling a sysem for he vehicle conrol sickers on January 1 s, 2010, i was possible o visually idenify he saus of a vehicle via a conrol sicker on he regisraion plae. A conrol sicker has previously indicaed ha vehicle has a valid insurance, paid a vehicle ax and passed a safey inspecion. This implies ha, wihou a regular police

10 10 Sherzod Yarmukhamedov conrol, i will be difficul o visually idenify wheher a vehicle in raffic has an acive or inacive license. Neverheless, in accouning for he vehicle park size in raffic, we include only he acive-license vehicles. I is expeced ha his variable (VEHP) is posiively associaed wih raffic faaliy since more vehicles in raffic increases he probabiliy of an acciden. As unemploymen level increases, he economic condiion of an individual worsens which may decrease affordabiliy of owning and driving a vehicle. The less reliance on a vehicle usage should decrease he probabiliy of raffic faaliies. These argumens are in line wih he evidences from Paryka (1984), Evans and Graham (1988) and Welki and Zlaoper (2007). Therefore, he unemploymen rae variable (UNEMP), measured in percenage, is expeced o have a negaive relaionship wih raffic faaliies. Road safey requiremens adoped via legislaions may conribue o a safer driving. Under he safey requiremens we imply enforcing he speed limi, alcohol inoxicaion while driving, sea bel law ec. The enforcemen of laws should reduce he number of raffic accidens, and hence faaliies on he road. As he enforcemen measures, Welki and Zlaoper (2007) use he arress for speeding, arress for no wearing sea bels and arress for drunk driving. We proxy hese raffic offences wih he number of revoked driving licenses due o he speeding (TROFF1) and drunk driving (TROFF2). I is expeced ha srenghening he law enforcemen would reduce raffic faaliies. High gasoline prices may lead o a decline in demand for vehicle use, which may decrease he probabiliy of raffic acciden, and hence faaliies. However, an increase in he price of gasoline may induce he vehicle drivers o swich o he moorcycles as an alernaive mode of ransporaion, so ha here migh be a decrease in he number of faaliies for passenger cars bu he faaliy rae migh increase for moorcycliss (Hya e al., 2009; Grabowski and Morrisey, 2004). The average price of 95 ocane gasoline is used o accoun for he fuel price (FUEL), SEK, curren prices. I is expeced ha he effec of he changes in fuel price is uncerain a priori. Cerain unmeasured, hus omied effecs ha are no in he model are expeced o improve he raffic safey over ime. These effecs migh be he sae of driver skill, qualiy of highway, privae demand and supply of he improved vehicle design, qualiy of he healh care, echnical advancemen in vehicle mainenance ec. (Pelzman, 1975). To accoun for he above described effecs, a ime rend variable (TREND) is included in he model, which equals one in year 1992 and increases by a year in he subsequen years. Many sudies have shown a negaive and significan associaion beween a rend variable and raffic faaliy rae, implying ha hese effecs are causing a downward rend in raffic faaliies over ime (Pelzman, 1975; Crandall and Graham, 1984; Welki and Zlaoper, 2007; Law e al., 2011). Therefore, i is expeced ha he impac of rend on raffic faaliies would be negaive.

11 Deerminans of Traffic Faaliies in Sweden Descripive saisics Descripive saisics are presened in Table 5.1. According o he able, he average number of annual raffic faaliies is abou 533. The mean income level is SEK per capia, while he average number of vehicle in raffic is The mean proporion of populaion possessing driving license is and , for young and old drivers respecively. As we may noe, he proporion of elderly drivers is as wice large as younger drivers. The average number of revoked driving licenses due o he speeding and drunk driving are and 7007, respecively. I seems ha he prevailing reason for driving license revocaions is he speeding offence. Finally, he average price of gasoline is SEK 9.4, and he mean rae of unemploymen during 17 years is 7.21%. Table 5.1: Descripive saisics ( ) Mean Sd. deviaion Min Max Traffic faaliy Income level Vehicle park size Populaion aged Populaion aged 67 and over Traffic offence: speeding Traffic offence: drunk driving Fuel price Unemploymen Resuls This secion presens he resuls of uni roo and mulicollineariy ess. Aferwards, based on he resuls of hese ess, we specify several models and provide he corresponding esimaion resuls. 6.1 Uni roo es According o he visual inspecion (Appendix 1), he disribuion of variables appears o be dissimilar, hus he models for esing a uni roo are specified individually for each variable. Traffic faaliy shows a rend and follows a random walk wih downward drif (Fig A.1), hus he corresponding model is: y 0 y 1 (6.1.1) where, y y y 1 is he firs difference of ime series, 0 is an inercep,

12 12 Sherzod Yarmukhamedov y 1 is a lag (firs) of variable, is a random disurbance wih mean zero and 2 consan variance v. Income level, populaion of old drivers, vehicle park and fuel price grow over ime exhibiing a ime rend (Fig.A.2), herefore he appropriae esing model is: y y v (6.1.2) where, is a ime rend. Populaion of young drivers, raffic offences for speeding and drunk driving, as well as unemploymen rae exhibi no definie rend and follows a random walk (Fig. A.3), hus he model for esing a uni roo is 3 : y y 1 (6.1.3) The saionariy requires he condiion where in all above presened equaions 0, oherwise he ime series are said o be nonsaionary. We es H : 0 agains H : 0. A % significance level, he null 0 1 hypohesis of a uni roo is rejeced if DF c, where DF c a criical se( ) value for he Dickey-Fuller es. The uni roo es resuls are presened in Table 6.1, which sugges ha he raffic faaliy rae is inegraed of order zero ( ln( FATAL ) ~ I(0) ), i.e. saionary, whereas oher variables are nonsaionary. By aking he firs differences of nonsaionary ime series, we obain saionary variables and hen use in he model esimaion. Table 6.1: Regression resuls of a uni roo es Coefficien Sd. error τ-saisic 5% criical value Traffic faaliy Income level Vehicle park size Populaion aged Populaion aged 67 and over Traffic offence: speeding Traffic offence: drunk driving Fuel price Unemploymen See Wooldridge (2013), Hill e al. (2001) and Gujarai (2004).

13 Deerminans of Traffic Faaliies in Sweden Mulicollineariy es A correlaion marix for he explanaory variables is presened in Table 6.2. Noe ha a rend variable is highly correlaed wih he variable for young populaion (0.89) and unemploymen rae is highly correlaed wih vehicle park (0.78) and income level (0.72). Therefore, some of hese variables should be excluded from he model. POP1 POP2 Table 6.2: Correlaion marix TRO TRO VEHP FF1 FF2 INC FUEL UNE MP POP1 1 POP VEHP TROFF TROFF INC FUEL UNEMP TREND TRE ND 6.3 Muliple regressions Considering he resuls of he uni roo and mulicollineariy ess, several model specificaions are considered. Model A ln( FATAL ) ln( INC ) ln( POP2 ) ln( VEHP ) ln( TROFF 1 ) ln( TROFF 2 ) 6 FUEL 7TREND In his model specificaion, he variables for young populaion and unemploymen are excluded, due o he srong correlaion beween he variables for: young populaion and ime rend; unemploymen rae and income level; unemploymen rae and vehicle park. Model B ln( FATAL ) ln( POP2 ) UNEMP ln( TROFF 1 ) ln( TROFF ) 5 FUEL 6TREND In his model, he excluded regressors in Model A are now included insead of he income level and vehicle park, i.e. he unemploymen rae is included in his model, while he income level, young populaion and vehicle park are excluded. 3 4

14 14 Sherzod Yarmukhamedov Model C ln( FATAL ) ln( INC ) ln( POP1 ) ln( POP2 ) ln( VEHP ) ln( TROFF 1 ) 6 ln( TROFF 2 ) 7 FUEL Unemploymen rae and rend variables are excluded in his model, due o he high correlaion beween he variables for: young populaion and ime rend; unemploymen rae and income level; unemploymen rae and vehicle park. Model D ln( FATAL ) ln( POP1 ) ln( POP2 ) UNEMP ln( TROFF 1 ) ln( TROFF 2 ) 6 FUEL Income level and vehicle park are replaced wih unemploymen rae, while a rend variable is replaced wih he young populaion variable due o he high correlaion. Regression esimaes of he above specified models are presened in Table Table 6.3: Regression resuls of raffic faaliy models Model A Model B Model C Model D Income level 1.624*** (0.786) (1.153) Populaion aged * (0.928) * (1.107) Populaion aged 67 and over (1.915) (1.712) (2.821) (3.191) Vehicle park size (2.289) ** (3.404) Traffic offence: speeding (0.137) (0.132) (0.201) (0.248) Traffic offence: drunk driving 0.626** (0.248) 0.633** (0.239) (0.336) (0.409) Fuel price < (0.000) <0.001 (0.000) < (0.000) <0.001 (0.000) Unemploymen *** (0.013) (0.024) Trend * (0.004) * (0.004) Consan 6.552* (0.086) 6.627* (0.062) 6.303* (0.112) 6.277* (0.096) Adjused R AIC ***, **, * Significan a 1%, 5% and 10%, respecively. Sandard errors are in parenheses.

15 Deerminans of Traffic Faaliies in Sweden 15 A choice of he model specificaion among four alernaives migh be based on he Adjused R-squared, where a model wih he highes value is seleced, i.e. Model B should be preferred. Anoher formal crierion o selec a model is Akaike s Informaion Crierion (AIC), where a model wih he lowes AIC should be seleced. The AIC values for four models sugges he selecion of Model B, herefore his model is used o explain he incidence of raffic faaliies. Esimaion resuls for Model B (only saisically significan coefficien esimaes are discussed) sugges ha a 10% increase in driving license revocaions due o he drunk driving increases raffic faaliies by 6.3%, all else equal. An increase in unemploymen rae by 1% leads o a decrease in raffic faaliies by 0.026%. A rend variable suggess ha he annual improvemen of driving skill, vehicle design, highway qualiy, healh care qualiy and oher insiuional road safey measures decrease raffic faaliies by 0.03%. 6.4 Resul discussion A uni roo es indicaes ha he raffic faaliy rae is saionary ime series ha follows a random walk wih downward drif, while he oher variables are differenced o urn hem ino saionary ime series. A mulicollineariy es of explanaory variables shows ha a ime rend variable is highly correlaed wih a variable for young populaion aged who has a driving license. I is also revealed ha he unemploymen rae has a srong correlaion wih he vehicle park and income level variables. Taking ino accoun he resuls of he uni roo and mulicollineariy ess, four models are specified. In consrucing hese models, we exclude highly correlaed regressors and replace he level (nonsaionary) variables wih heir firs differences. A choice of he model is based on he values of he adjused R-squared and AIC, where he seleced model is consruced from such variables as he populaion of elderly drivers, unemploymen rae, raffic offences due o he speeding and drunk driving, fuel price and ime rend o explain he incidence of raffic faaliies. Resuls indicae ha an increase in driving license revocaions due o he drunk driving increases raffic faaliies. The expecaion was ha driving license revocaions should decrease he number of faaliies on he road by expelling he drivers ha violae he raffic safey laws, especially, when he cause of infringemen is alcohol inoxicaion 4. However, Welki and Zlaoper (2007) find a similar resul and argue ha afer faal accidens, police may enforce he raffic safey regulaion more sricly. Our daa suppor his argumen, i.e. a sudden increase in he number of raffic faaliies (9.23%) in 1999 is followed a year laer by a sharp increase in he number of driving license revocaions due o he drunk driving (11.68%). This implies ha afer raffic accidens, police underake safey 4 Sen and Campbell (2010) find ha adminisraive license revocaion policies reduce raffic faaliies.

16 16 Sherzod Yarmukhamedov measures by exering more sringen conrol on he roads. Anoher finding indicaes ha an increase in unemploymen rae is associaed wih a decline in raffic faaliies. Worsening of he purchasing power due o he increased unemploymen rae has he negaive influence on vehicle ownership. Consequenly, fewer vehicles in raffic decreases he probabiliy of raffic faaliies. This resul is consisen wih Paryka (1984), Evans and Graham (1988), as well as Welki and Zlaoper (2007). Moreover, we also find ha here are some unmeasured facors ha improve raffic safey over ime. The facors ha conribue o he decline in raffic faaliy rae migh be he improvemen of driving skill, vehicle design, highway qualiy, medical care and oher raffic safey measures, which are in line wih he findings of Pelzman (1975), Crandall and Graham (1984), Welki and Zlaoper (2007) and Law e al. (2011). 7 Conclusion This sudy empirically analyzes he incidence of road raffic accidens in Sweden from he perspecive of differen socio-economic facors. In he presence of he varying performance of road raffic sysems in differen counries and conflicing findings of previous research, he purpose of his sudy is o deermine he socio-economic facors ha may explain he incidence of raffic faaliies. Analysis uses he Swedish annual daa over he period from 1992 o 2009, where a raffic faaliy rae is modelled as a funcion of populaion of elderly drivers, raffic offence due o he drunk driving, unemploymen rae, fuel price and ime rend. Afer conducing a uni roo and mulicollineariy ess, a raffic faaliy model is consruced. We find ha an increase in driving license revocaions due o he drunk driving increases raffic faaliies. Despie he common belief ha more revocaions should decrease he number of raffic faaliies, our resul suggess he opposie. This migh be explained by srenghening he conrol by police as a consequence of he pas faal accidens. Moreover, worsening of he economic siuaion due o an increase in unemploymen rae conribues o a reducion in raffic faaliy rae. Finally, some unmeasured facors such as he improved driving skill, vehicle design, highway qualiy, medical care and oher raffic safey measures significanly decrease raffic faaliies. Resuls of his sudy are in line wih oher findings in he lieraure, and our empirical sudy invesigaed he incidence of raffic accidens associaed wih faaliy based on aggregae daa, whereas a separae analysis for specific groups on micro level daa migh shed a beer ligh o he invesigaed issue which would serve as a reason for furher research. This research disinguishes iself from he previous sudies in hree respecs. Many ime series sudies addressing he issue of raffic faaliies do no menion or recognize he poenial problem of nonsaionariy. Presen sudy invesigaes he naure of nonsaionariy of he variables and parsimoniously specifies a model for

17 Deerminans of Traffic Faaliies in Sweden 17 each variable o es for uni roos o avoid he problem of spurious regressions. Moreover, we use several variables ha are newly inroduced by recen sudies in raffic safey lieraure. Furhermore, his is he firs sudy analyzing he impac of socio-economic facors on raffic faaliy rae for he case of Sweden. References [1] WHO, World repor on raffic injury prevenion, The World Healh Organizaion, (2004), [2] N. Anbarci, M. Escaleras, and C. Regiser, Traffic faaliies and public secor corrupion. KYKLOS, 59(3), (2006), [3] D. Bishai, A. Quresh, P. James and A. Ghaffar, Naional road casualies and economic developmen, Healh Economics, 15, (2006), [4] E. Kopis and M. Cropper, Traffic faaliies and economic growh. The World Bank Policy Research Working Paper 3035, (2003). [5] R. Noland, Traffic faaliies and injuries: he effec of changes in infrasrucure and oher rends, Acciden Analysis and Prevenion, 35(4), (2003), [6] T.H. Law, R. Noland, R. and A. Evans, The sources of he Kuznes relaionship beween road faaliies and economic growh, Journal of Transpor Geography, 19(2), (2001), [7] A. Trawe, P. Marase and U. Persson, Inernaional comparison of coss of a faal casualy of road accidens in 1990 and 1999, Acciden Analysis and Prevenion, 34, (2002), [8] S. Blows, S. Ameraunga, R. Ivers, S. Lo and R. Noron, Risky driving habis and moor vehicle driver injury, Acciden Analysis and Prevenion, 37, (2005), [9] A. Tarko, Modelling drivers speed selecion as a rade-off behavior, Acciden Analysis and Prevenion, 41, (2009), [10] S. Levi and J. Porer, How dangerous are drinking drivers? Journal of Poliical Economy, 109(6), (2001), [11] B. Sen and C. Campbell, Alcohol prevalence, alcohol policies, and child faal injury raes from moor vehicle crashes, Conemporary Economic Policy, 28(3), (2010), [12] S. Islam and F. Mannering, Driver aging and is effec on male and female single-vehicle acciden injuries: some addiional evidence, Journal of Safey Research, 37, (2006), [13] M. Simoncic, Road accidens in Slovenia involving a pedesrian, cyclis or moorcyclis and a car, Acciden Analysis and Prevenion, 33, (2001), [14] Y. Mori and M. Mizohaa, Characerisics of older road users and heir effec on road safey, Acciden Analysis and Prevenion, 27(3), (1995),

18 18 Sherzod Yarmukhamedov [15] L. Beck, A. Dellinger and M. O Neil, Moor vehicle crash injury raes by mode of ravel, Unied Saes: using exposure-based mehods o quanify differences, American Journal of Epidemiology, 166(2), (2007), [16] R. Elvik, and T. Vaa, Handbook of Road Safey Measures, Elsevier, Unied Kingdom, [17] Z. Wong, C. Chong, B. Tai and G. Lau, A review of faal road raffic accidens in Singapore from 2000 o 2004, Annuals Academy of Medicine, 38, (2009), [18] D. Clark and B. Cushing, Rural and urban raffic faaliies, vehicle miles, and populaion densiy, Acciden Analysis and Prevenion, 36, (2004), [19] E. Hya, R. Griffina, L. Rue and G. McGwin, The associaion beween price of regular-grade gasoline and injury and moraliy raes among occupans involved in moorcycle- and auomobile-relaed moor vehicle collisions, Acciden Analysis and Prevenion, 41(5), (2009), [20] D. Grabowski and M. Morrisey, Gasoline prices and moor vehicle faaliies, Journal of Policy Analysis and Managemen, 23(3), (2004), [21] F. Wilson, J. Simpson, and P. Hilsenrah, Gasoline prices and heir relaionship o rising moorcycle faaliies, , American Journal of Public Healh, 99(10), (2009), [22] T. Wenzel, M. Ross, The effecs of vehicle model and driver behavior on risk, Acciden Analysis and Prevenion, 37(3), (2005), [23] L. Lam, Facors associaed wih young drivers car crash injury: comparisons among learner, provisional, and full licensees, Acciden Analysis and Prevenion, 35(6), (2003), [24] Vägverke, Safe raffic: Vision Zero on he move. The Swedish Transpor Adminisraion, [25] R. Elvik, M. Kolbensved, B. Elvebakka, A. Hervikb and L. Bræinc, Coss and benefis o Sweden of Swedish road safey research, Acciden Analysis and Prevenion, 41(3), (2009), [26] J.M. Wooldridge, Inroducory Economerics: A Modern Approach, Fifh ediion, Cengage Learning, Unied Kingdom, [27] R.C. Hill, W.E. Griffihs and G.G. Judge, Undergraduae Economerics, Second ediion, John Willey & Sons, USA, [28] S. Pelzman, The effecs of auomobile safey regulaion, The Journal of Poliical Economy, 83(4), (1975), [29] T.J. Zlaoper, Models explaining moor vehicle deah raes in he Unied Saes, Acciden Analysis and Prevenion, 21(2), (1989), [30] T.J. Zlaoper, Deerminans of moor vehicle deahs in he Unied Saes: A cross secional analysis, Acciden Analysis and Prevenion, 23(5), (1991),

19 Deerminans of Traffic Faaliies in Sweden 19 [31] S. Paryka, Simple models of faaliy rends using employmen and populaion daa, Acciden Analysis and Prevenion, 16(3), (1984), [32] W. Evans, and J.D. Graham, Traffic safey and he business cycle, Alcohol, Drugs, and Driving, 4(1), (1988), [33] A.M. Welki, and T.J. Zlaoper, The impac of highway safey regulaion enforcemen aciviies on moor vehicle faaliies, Transporaion Research Par E, 43, (2007), [34] R.W. Crandall and J.D. Graham, Auomobile safey regulaion and offseing behavior: Some new empirical esimaes, The American Economic Review, 74(2), (1984), [35] D.N. Gujarai, Basic economerics, Fourh ediion, The McGraw-Hill Companies, New York, 2004.

20 20 Sherzod Yarmukhamedov Appendix Figure A.1: Economic ime series: a random walk wih drif Figure A.2: Economic ime series: a ime rend

21 Deerminans of Traffic Faaliies in Sweden 21 Figure A.3: Economic ime series: a random walk