INTERNATIONAL JOURNAL Of ACADEMIC RESEARCH Vol. 2.. 5. September, 2010 MOTORCYCLE ACCIDENT MODEL FOR SEVERITY LEVEL AND COLLISION TYPE Amelia Kusuma Indriastuti, Harnen Sulistio Department of Civil Enginering, Faculty of Engineering, Brawijaya University, Malang 65145, (INDONESIA) ABSTRACT Motorcycle accident becomes the most frequent incident in traffic accident. To increase motorcyclists safety, the fact of safety problems of motorcyclist is needed as the first step to develop the appropriate action program. This paper is aimed to develop a model that shows the relationship between motorcycle accident collision type and the fatal casualties generated. The model is developed by the Generalized Linear Modeling (GLM) method. The models (with 10% confidence level) show that for the number of rear-end collision between 1 to 5 accident per annum, the fatal casualties will increase up to 27.63%, compared to the number of rear-end collision less than 1 per annum. Whilst, the number of side collision between 1 to 5 per annum will increase the number of fatal casualties up to 21,65% per annum. The number of fatal casualties will raised significantly in a head-on collision, where the fatal casualties will increase up to 43.91%. Key words: accident model; fatal casualties; collision type; Generalized Linear Modeling 1. INTRODUCTION Motorcycle has become a type of transportation mode dominating the traffic in Indonesia. In 2007, motorcycle population was about 78,3% of the total number of motorized vehicle or about 37 million units (Lubis, 2008) [9]. Motorcycle growth rate is about 21.06%. According to the existing trend, it is predicted that the number of motorcycles will increase. As a result, the potency of motorcycle-involved accidents will also increase. To date, the involvement of motorcycle in accidents contributed the most number of fatalities, that is, 75% of fatalities were motorcyclists (GRSP, 2007) [4]. The number and the types of vehicles involved in accident in Indonesia is depicted in Figure 1. It is fact that the number of motorcycle-involved accident was the most among other types of vehicles involved in accident. In 2004, the number of motorcycle-involved accidents was 14,233 cases (of 25,969 cases of total accident or 54.79%) in Indonesia. Lubis (2008) even stated that in 2007, the number of motorcycle-involved accidents increased to 65% [9]. Fig. 1. The number and the types of vehicles involved in accident in Indonesia During 2003, there were 75,000 fatalities and 4.7 million injuries as a result of traffic accident in ten of ASEAN countries. The financial loss in the event of traffic accident was sufficient high, that is, USD 15 billion or equaled to 2.2% of gross domestic product (GDP) of ASEAN countries. Indonesia suffered the most suffer with estimated financial loss about USD 6.03 billion (2.91% of GDP). If there is no serious safety improvement in the next five years, it is predicted that 385,000 fatalities and 24 million injuries will be occured ASEAN countries with estimated financial loss about USD 88 billion (ADB, 2004). It means that the impact of the traffic accident has become a ral threat for human being and be a serious problem for all of the world [2]. In order to fulfill the aforementioned targets, the first important step to be done is to understand the characteristics of motorcyclist and the characteristics of motorcycle-involved accident. The different of socioeconomics characteristics and the behaviour of motorcyclist are the main considerations to identify the factors contributing traffic accidents, strategic steps in the form of policy and regulation about how to ride safely, safety campaign for motorcyclist, and so on. 210 www.ijar.lit.az
INTERNATIONAL JOURNAL Of ACADEMIC RESEARCH Vol. 2.. 5. September, 2010 The big number of motorcycle population is one important characteristic of developing countries in Asia. The number of motorcycles in Indonesia is about 78.3% of the total number of all vehicles. This emerges a crucial problem in traffic safety as 65% of traffic accident in Indonesia in 2007 was motorcycle-involved accident. According to this fact, it is necessary to create an action program to increase the safety of motorcyclist. To date, the Indoesia goverment has launched several action programs to respond the aforementioned challange, although most of the programs were adopted from the programs developed by other countries and there were no effort to evaluate the running action programs. Based on this condition, it is important to find out the source of safety problems of motorcyclist as the first step to develop the appropriate action program for the existing problems. The proposed action programs will need an evaluation pertaining to the effectiveness of each action program implemented. To support this, it needs an academic study that will be a reference for decision making. The motorcyclist safety is a part of land safety. The motorcycle population increases together with the increase of traffic motorcycle-involved accident. The accident occurred could cause by several factors, such as by human (motorcyclist or other road users), road structure, traffic, vehicles and environment. Indriastuti (2006) identified that 55.4% of accidents in Malang regency during years 2000 to 2005 was motorcycle-involved accident [8]. Harnen et. al (2003, 2004) revealed that the factors caused the motorcycle accidents at intersections in Malaysia were traffic condition, speed, lane width, number of lane, shoulder width and land use at intersection legs [6,7]. Suraji et. al (2007) stated that human factor was the main factor in accident occurrences, while other factors were not significant [16]. Ueda et. al (2007) did a research about the accidents and their causes and found that tiredness of the motorcyclists was the significant factor for the traffic accident [19]. Ahmad et. al (1999) did a research on motorcyclist knowledge, attitude and practice in Malaysia which is related with traffic safety [1]. The thinking basis of this study was the fact that the motorcycle-involved accident was the largest in number, and human factor dominates the cause of the accident. By understanding of characteristics of motorcyclist and motorcycle-involved accidents, it is expected that a list of action program to improve motorcyclist safety based on the motorcyclist socio-economic characteristics could be developed in study area. The types of action programs were selected based on public aspiration, either the motorcyclist themselves or others, such as health and educational institutions, and so on. The public aspiration in improving the motorcyclist safety was going well with the safety improvement roadmap in Indonesia. It is expected that the implementation of the action programs that was effectiveness-proved in this research could be a good example for other region with similar characteristics with the study area. This study is meant to obtain the fatality accident model due to motorcycle accident collision type. From this model, it can be seen the impact of each accident collision to the fatal casualties generated. Furthermore, this model is expected to give a contribution in the implementation of the action programs. The development of statistical models to explain factors contributing to accidents has been a subject of numerous traffic accident studies. Generally, such studies aimed to establish statistical models relating traffic accidents to various measures of traffic flow, site characteristics and road geometry. The statistical models allowed the accidents to be predicted and so that the appropriate countermeasures or treatments could be established. Early studies on traffic accident modeling used a conventional linear regression approach, which was based on the assumption of normal error distribution with a constant variance. Although this approach contributed significantly to the knowledge of traffic accident modeling, the assumptions used in this approach seem to be inappropriate due to the characteristics of traffic accidents. The conventional linear regression lacked distributional property to adequately describe such accident characteristics. Therefore, the use of more general form of conventional linear regression, the generalized linear modeling (McCullagh and Nelder, 1989) is more appropriate in modeling traffic accidents [10]. This preferred approach is available in computer programs such as GLIM, SAS, LIMDEP and GENSTAT. Earlier studies reported the use of generalized linear modeling approach to develop statistical models relating traffic accidents to traffic flow and road characteristics. They used either the cross-sectional or the time series analysis to develop the models (Griebe and Neilson, 1996; Tarko et al., 1999; Vogt and Bared, 1998; Vogt, 1999; Rodriquez and Sayed, 1999; Radin et al., 1995; Radin, 1996; Bauer and Harwood, 2000; Saied and Said, 2001; Taylor et al., 2002) [5, 17, 20, 21, 14, 12, 13, 3, 15, 18]. In generalized linear modeling, a statistical model consists of two components (McCullagh and Nelder, 1989; NAG, 1994), the systematic and random components [10, 11]. The systematic component describes the way in which the independent (explanatory or covariate) variables combine together to explain the variation of the dependent (response) variable, and the random component describes the error term. The linear combination of the explanatory variables is called linear predictor. There are three main components of a model in generalized linear modeling: 1. The random component: Error or probability distribution f(y) which has a mean. 2. The systematic component: Linear predictor or linear regression function ( ) For n explanatory variables: = = 0 + 1 x 1 + + n x n (1) where: 0 (sometimes called intercept) and i are parameters to be estimated x i is covariates x 1, x 2, x n 3. Link function or parameter transformation (g), = g( ). This function links the linear predictor (systematic component) to the mean (random component). In conventional linear regression analysis, the assumptions are: Baku, Azerbaijan 211
INTERNATIONAL JOURNAL Of ACADEMIC RESEARCH Vol. 2.. 5. September, 2010 1. The probability distribution of the dependent variable y is normal, N (, 2 ), with mean and constant variance 2. 2. The linear predictor is: = = 0 + 1 x 1 + + n x n (2) 3. The link function is identity (i.e. no transformation) 2. METHODS The determination of a motorcycle-involved accident model used a method of Generalized Linear Modeling (GLM). An exponential model is commonly employed to represent this method with assumption that the characteristic of the data is in Poisson distribution. The general equation of this proposed model is expressed by: MCA = k EXP ( ACT) (3) in which MCA is the number of motorcycle-involved accident victims; ACT is the types of collision (i.e. headon collision, rear-end collision, side collision and run-off-road collision); and k, β, are the coefficient of independent variables. 3. RESULTS AND DISCUSSION Model Description Motorcycle accident model in this study is a model that is a function of types of collision and the severity of the victims. Types of collision considered are rear-end collision (TB), head-on collision (TD) and side collision (TS). In addition, two severities of victims are considered, i.e. severe injury (LB) and fatal injury (MD). In developing a motorcycle-involved accident model, Generalized Linear Modeling (GLM) was used. An exponential model is commonly employed to represent this method with assumption that the characteristic of the data is in Poisson distribution. The general models used are as follows. MD = k 1 EXP ( 1 ACT) (4) LB = k 2 EXP ( 2 ACT) (5) in which MD = the number of fatal injuries as a result of motorcycle-involved accident LB = the number of severe injuries as a result of motorcycle-involved accident ACT= the types of collision that can be categories into several sub-variables as follows TB (1) = the number of rear-end collision less than 1 per annum TB (2) = the number of rear-end collision between 1 to 5 per annum TB (3) = the number of rear-end collision more than 5 per annum TD (1) = the number of head-on collision less than 1 per annum TD (2) = the number of head-on collision between 1 to 5 per annum TD (3) = the number of head-on collision more than 5 per annum TS (1) = the number of side collision less than 1 per annum TS (2) = the number of side collision between 1 to 5 per annum TS (3) = the number of side collision more than 5 per annum k i, i = the coefficient of independent variables The analysis conducted in this study was univariate analysis. This is because there is no relationship between types of collisions. Univarate analysis was performed to know the contribution of each independent variable (see Tables 1 and 2). Table 1 shows that almost all of independent variables give a significant influence on dependent variables at 95% significance level (α = 0.05). At 90% significance level, all independent variables give significant influence on dependent variables. With the addition of another variable, such as speed, the increase of the number of head-on, rear-end and side-collisions will contribute to the increase of the number of fatal injuries. Table 1. The results of univariate analysis for fatal injuries (MD) Explanatory Coefficient Standard t-value = 0.05 = 0.10 Variable Error Constant 1.276 0.0693 3.52 Yes Rear-end collision (2)* 0.244 0.127 1.92 Yes Rear-end collision (3)* 1.414 0.229 6.18 Yes Constant 1.182 0.0807 2.07 Yes Head-on collision (2)* 0.364 0.115 3.18 Yes Head-on collision (3)* 1.625 0.3 5.42 Yes Constant 1.296 0.075 3.46 Yes Side collision (2)* 0.196 0.116 1.7 Yes Side collision (3)* 1.35 0.325 4.15 Yes Remarks: * variable coefficients (2) and (3) are ratios against variable coeffiient (1) 212 www.ijar.lit.az
INTERNATIONAL JOURNAL Of ACADEMIC RESEARCH Vol. 2.. 5. September, 2010 Table 2. The results of univariate analysis for severe injuries (LB) Explanatory Coefficient Standard t-value = 0.05 = 0.10 Variable Error Constant 1.188 0.127 1.36 Rear-end collision (2)* -0.086 0.315-0.27 Rear-end collision (3)* 1.214 0.257 4.72 Constant 1.056 0.148 0.37 Head-on collision (2)* 0.41 0.228 1.8 Head-on collision (3)* 1.267 0.298 4.26 Constant 1.042 0.17 0.24 Side collision (2)* 0.362 0.222 1.63 Side collision (3)* 1.568 0.359 4.36 Remarks: * variable coefficients (2) and (3) are ratios against variable coeffiient (1) Table 2 indicates that the results of the univariate analysis at = 0.05 for a relationship between the types of collision and the number of victims are not significant enough. The same results were found even when the level of significance was reduced to 90% ( = 0.10). Fatality Accident Model due to Motorcycle Accident Type Using the proposed model, the final models of motorcycle accident based on the severity of the victims are as follows. (6) A good model should be able to estimate closely the observation data. Therefore, it is necessary to evaluate the models by means of the deviation between estimated and observed values produced. Figure 1 to Figure 3 show that the estimated values of the model is sufficiently close with the observed ones (at = 0.10). It means that the resulted model could represent the existing condition. In those figures, it can be seen that the comparison between estimated and observed values for each collision type. (7) (8) Number of Fatalities per annum observation model estimation Fig. 1. Comparison between estimated and observed values for rear-end collision (TB) Baku, Azerbaijan 213
INTERNATIONAL JOURNAL Of ACADEMIC RESEARCH Vol. 2.. 5. September, 2010 Number of Fatalities per annum observation model estimation Fig. 2. Comparison between estimated and observed values for head-on collision (TD) Number of Fatalities per annum observation model estimation Fig. 3. Comparison between estimated and observed values for side collision (TS) From Figure 1 to Figure 3, it seems that the model could well estimate the observed values at 90% significant level. Model Interpretation The resulted accident model indicated that the types of collision affected the number of fatal injuries. The following table summarizes the comparison of the number of fatal injuries for three different types of collision. Table 3. Comparison of the number of fatal injuries for different types of collisions Types of collision Number of fatal injuries based on number of collisions (Collision 1 5) against (collision < 1) (Collision > 5) against (collision < 1) Rear-end collision 27.63% higher 3 times higher Head-on collision 43.91% higher 4 times higher Side collision 21.65% higher 3 times higher 214 www.ijar.lit.az
INTERNATIONAL JOURNAL Of ACADEMIC RESEARCH Vol. 2.. 5. September, 2010 Table 3 indicates that among the three types of collision, head-on collision caused the most number of fatal injuries. The head-on collisions in a road segment are commonly occurred in the following cases. a. undivided roads, in which the collision of two vehicles from different direction were occurred at the middle of the road. The main causes of the accident were marker violations and speeding. b. lane direction violation, especially by motorcyclist who dislike doing u-turn on appropriate places. c. Weakness in law enforcement contributes on the occurrence of many violations and some of them will be ended by accidents. 4. CONCLUSION The fatality model shows that for the number of rear-end collision between 1 to 5 accident per annum, the fatal casualties will increase up to 27.63%, compared to the number of rear-end collision less than 1 per annum. Whilst, the number of side collision between 1 to 5 per annum will increase the number of fatal casualties up to 21,65% per annum, compared to the number of side collision less than 1 per annum. The number of fatal casualties will raised significantly in a head-on collision, where the fatal casualties will increase up to 43.91% compared to the number of side collision less than 1 per annum. REFERENCES 1. Ahmad S., Radin Umar, RS. 1999. Showcase of Road Safety Study An Engineering Approach. 2. Asian Development Bank (ADB). 2004. ASEAN Region Road Safety Strategy and Action Plan, Final Draft Report. Asian Development Bank. Manila. 3. Bauer, K.M., Harwood, D.W. 2000. Statistical Models of At-grade Intersection Accident-Addendum. Publication. FHW A-RD-99-094. Federal Highway Administration. McLean, New Virginia 4. Global Road Safety Partnership. 2007. Cascading the Worl Report The ASEAN Experience. 5. Griebe, P., Nielsen, M.A. 1996. Safety at Four-armed Signalized Junction Situated on Roads with Different Speed Limits. Proceedings of the Conference on Road Safety in Europe. Birmingham, UK. 6. Harnen, S., R.S. Radin Umar, S.V. Wong, W.L. Wan Hashim. 2003. Motorcycle crash prediction model for non-signalized intersections. Journal of IATSS Research. Vol 27, 2. page 58-65. 7. Harnen, S., R.S. Radin Umar, S.V. Wong. 2004. Development of prediction model for motorcycle crashes at signalized intersection on urban road in Malaysia, Journal of Transportation and Stastistic, Vol 7 3. page. 27-39. 8. Indriastuti, A.K, Ambarwati, L, Setiadji, B.H., 2007. Preliminary Study of Accident Characteristics in Malang District, Indonesia. Proceeding of 5th Asia Pacific Conference on Transportation and Environmental Engineering. Singapore, Singapore. 9. Lubis, H.A.R.S., 2008, Pertumbuhan Sepeda Motor dan Dampaknya Bagi Transportsasi Perkotaan, Jurnal Transportasi, Vol 8. 3, FSTPT, hlm. 187-24. 10. McCullagh, P., Nelder, J.A. 1989. Generalized Linear Model, 2 nd Edition. Chapman and Hall. New Jersey 11. Numerical Algorithm Group (NAG). 1994. The GLIM System, Release 4 Manual, 2 nd Edition. Clarendon Press. Oxford 12. Radin Umar, RS., Mackay, GM., Hills9, BL. 1995. Preliminary Analysis of Exclusive Motorcycle Lanes along The Federal Highway F02 in Shah Alam, Malaysia. IATTS Research Vol. 19. 2. page 93-98. 13. Radin Umar, RS. 1996. Accident Diagnostic System With Special Reference to Motorcycle Accidents in Malaysia. Ph.D Thesis. University of Birmingham, England. 14. Rodriguez, L.P., Sayed, T. 1999. Accident Prediction Models for Urban Unsignalized Intersection in British Columbia. The 78 th Annual Meeting of Transportation Research Board. Washington DC 15. Saied, A.M., G.M. Said. 2001. A General Linear Model Framework for Traffic Conflicts at Uncontrolled Intersection in Greater Cairo. Proceedings of The Conference, Traffic Safety on Three Continents. Moscow, Russia 16. Suraji, A; N. Tjahjono, S. Harnen, 2007, Pemodelan Kecelakaan Pengendara Sepeda Motor Pada Kawasan Perkotaan, Laporan Akhir Penelitian PHB Dikti, Fakultas Teknik Universitas Widyagama Malang. (Tidak dipublikasikan). 17. Tarko, A.P., Eranky, S., Sinha, K.C., Scienteie, R. 1999. An Attempt to Develop Crash Reduction Factor Using Regression Technique. The 78 th Annual Meeting of Transportation Research Board. Washington DC. 18. Taylor, M., Kennedy, J.V., and Baruya, A. 2002. The Relationship Between Speed and Accidents on Rural Single-Carriageway Roads. Report TRL 511. Crowthorne, UK 19. Ueda M., A. Kondo, H. Matsumoto, H. Hayakawa; T. Nakata, 2005, Fatigue Analysis Based On Synthesis Of Psychological And Physiological Responses Measured Simultaneously In Follow-Up Driving, Journal of the Eastern Asia Society for Transportation Studies, Vol. 6, page 3325 3340. 20. Vogt, A., Barred, J.G. 1998. Accident Models for Two Lane Rural Road: Segment and Intersection. Report. FHW A-RD-98-133. Federal Highway Administration. McLean, Virginia 21. Vogt, A. 1999. Crash Models for Intersection: Four Lane by Two Lane Stop Controlled and Two Lane by Two Lane Signalized. Report. FHW A-RD-99-128. Federal Highway Administration. McLean, Virginia Baku, Azerbaijan 215