J. 0. Asalor Mechanical Engineering Department, University of Benin, P.M.B. 1154, Benin City, Nigeria E. A. Onibere Institute of Computer Science, University of Benin, P.M.B. 1154, Benin City, Nigeria (Received May 1984; revised August 1984) An offshoot of an existing general model of road traffic accidents (M RTA) is presented. The general M RTA contains an unwieldy number of terms for practical purposes. A reduction in the number of terms becomes necessary to render the model suitable for numerical computa- tions. The method used to achieve this through grouping primary causes is shown. Some potential applications of the model are also outlined. Computations performed using this model yielded results which com- pared favourably with available results from other practical sources. Key words: mathematical model, road accidents, interactions A general model of road traffic accidents (MRTA) that took account of all potential primary causes of traffic acci- dents was developed in a previous paper. Included in the model was each absent or malfunctioning component that could relate to the safe use of either road or vehicle; every irresponsible manoeuvre and failure to avert an accident by means of a normal safe manoeuvre when this option was available. While the general model provides a basic theoretical framework that applies to a wide range of traffic situa- tions, the expressions contain too many terms making solution difficult. The aim of this paper is to develop a simpler version of the general model that contains fewer terms and is more suited to computational purposes. This has been achieved in two stages. First, primary causes are grouped into a number of mutually exclusive sets each set being represented by a unifying cause: a root primary cause (r.p.c.). Expressions are then derived that correspond to those of the general model in terms of the r.p.c.s, and a further reduction of the number of terms is presented. Grouping accident causal variables Four main factors are responsible for the large number of terms that appear in the expressions of the general model: (1) The large number of interactions among vehicles in- cluded in the model, only a small proportion of which actually result in accidents. 0307-904X/85/042275-06/$03.00 0 1985 Butterworth & Co. (Publishers) Ltd (2) The large number of complex interactions among the primary causes. While the model took account of all the interactions, only a small proportion may actually lead to accidents. (3) The number of primary causes of accidents that appear explicitly in the general model is large. (4) A road accident is normally a rare event and the pro- cess of occurrence is very short compared with the inter- val between accidents on a road element. Use of a time- domain therefore leads to a large number of time intervals in which there is no accident. In this paper the number of variables in the model is first reduced. An examination of primary causes of road acci- dents reveals that they are all traceable to a relatively small number of common roots. Figure 1 shows a typical r.p.c. that has been decomposed to its primary causes. In between, there are a number of quasi-root primary causes. For the simpler model to retain the accuracy and generality of the original model, the r.p.c.s must meet the following con- ditions: (1) The risk variable that relates to a particular r.p.c. must fully represent the risk variables of all the associated pri- mary causes actina alone or in combination among them- selves. Let I be the total number of primary causes and K the corresponding number of r.p.c.s such that: Rr,j, . . . , R,,,j relate to @r,i Appl. Math. Modelling, 1985, Vol. 9, August 275 A model of road accidents: J . 0. Asalor and E. A. Onibere Root primarv Quasi-root prrmary causes Prrmary causes, R,, , .I ,=l 2 34 5.,.. 1 Figure 7 A root primary cause decomposed to its primary causes Tab/e 7 Two typical groups of root primary causes of road traffic accidents Group one Group two 1 Vehicle condition 2 Road and its surroundings (including road signs and weather condition) 3 Drivers safety performance 4 Pedestrians and other road users safety performance 5 Circumstance 1 Vehicle condition 2 Road and its surroundings (including road signs and weather condition) 3 Human safety performance 4 Circumstance R ?I 1 +1,j, . . . > R n,,j relate to @a,j R . nk-l+l,j R nk, i relate to &j R ?I K_l+l,j> > R&i relate to @K,] Then dk,j can be obtained from the primary causes by: @k,j = F Ril,j + T F RilrjRilrj + . . . i,=tk i,tk i,= tk + i ztk i2$tk . . iy$tk Ri,,iRi2.i . . . Riysl I + Rtk.jRtk,l,jRtk+2..Rnk,j where: tk=nk__l+ 1 i,<iz<i3<...<iu @k, j and Ri,j represent the root and primary risk variables respectively, k and i denote the respective causes and i is the vehicle that is involved. (2) Interaction among Ri,js from, say three different r.p.c.s is represented by interaction that involves exactly the same three r.p.c.s. The above role of the r.p.c. is similar in some respects to that of common cause in reliability studies.2-4 Two sets of r.p.c.s are proposed in Table 1. The list is not exhaustive. Each r.p.c. represents a classical subsystem of a traffic system, for example in traffic engineering and controlsS6 and road traffic safety.7,8 Circumstance reflects the role of a traffic situation involving innocent drivers and pedestrians in accidents. Formulation of the equations By introducing r.p.c., the general model can be split into two parts. The part of most interest here relates an accident to the number of vehicles and the r.p.c. involved. The other part relates each r.p.c. to its primary causes and has been dealt with in the preceding section. The root risk variables @k, j where k is the r.p.c. andj is the vehicle affected is a binary with the following proper- ties: @k,j = 1 if there is an accident due to k alone = 0 otherwise (la) @k,,j @k,,j = 1 ifth ere is an accident due to r.p.c. kl and kz only = 0 otherwise (lb) @k,,j@k,,j . . @k,j = 1 if there is an accident due to r.p.c. k,,k,,...,k,only = 0 otherwise (lc) Following the procedure of reference 1 the effective risk of an accident in the time interval dt(m) as it affects j becomes: kff,j(m> = @ji(l, m>+ Gji(2, nz> 2 @x,,2(2, m> x,=2 + @(3,m) g f h Qxi(3, m) + . . . x,=2 x,=2 i=2 + @j(r, WI) F F . . F A cbxik m> + . . x,=2 x,=2 x,52 i=2 + @j(N ml i f . . . f ii @&Y m> (24 x,=2 x,=2 xN=2 i=2 wherejfxz#x3fx4f...#xN,or: &ff,j(m> = Gji(l, ml + f r=2 X { 4jCr, mi[ xc2. . . i ;I ~Xi(r, ml]} 2 x,=2 i=2 (2b) where N is the largest number of vehicles that could be in- volved in a single accident. Now: @j = @j(@l,j, @2,j, . . .j @K, j) (34 Eli = ~~i(~l,~i ~2,~i . . . ) ~K,Xi) WI where K is the total number of r.p.c. of accidents as it affects one vehicle. Then: &(Y, m) = risk of xi being involved in an r-vehicle accident by a single r.p.c. + any two r.p.c. + any three r.p.c. + + any S r.p.c. (4) 276 Appl. Math. Modelling, 1985, Vol. 9, August A model of road accidents: J . 0. Asalor and E. A. Onibere where S is the maximum actual number of r.p.c.s that could involve a single vehicle in an accident (S < K). Combining equations (3) and (4) and after simplification, one has: . . ,g, ,h, @kp,Xi(" m)} (1 s wherekl<k,<...<k,. Substituting for $x,(~, m) into equation (2b): &ff, j(m) (6) The following constraints hold for a short time interval: @,ff,j(m) G 1 Pa) ,f @k,,j (l , m> G 1 (7b) k, =1 5 { f . . . f ii @kp,j (l ,m)/~ 1 (7c) s=2 k,=l k,=lp=l Equations (S)-(7) contain many fewer terms than those of the general model. Following the analysis of reference 1, equations (5) and (6) can be summed over an infinitely large number of short time intervals dt(m), and the cumulative effective risk &n,i(cM) obtained. The effective risk @,n,i(cM) can be expressed as a function of the number of vehicles that are involved in an accident as: [.t2 .E2 j2 @Xicr2 cM)]] c8) r If the total vehicle distance travelled and the total vehicle exposure time are X and T respectively,*then the mean effective risk per unit vehicle distance, &rr, j, and per unit time of vehicle exposure &n. i, can be expressed as: &f, j = lim -G&f, j(CM) etc. T--t- T Equation (8) becomes: &ff, j = lim $j(l, CM) X+m x 1 +;{GjCr.CW[ f . . . f li r=2 x,=2 x,=2 i =2 x @5& CM) Ill (9b) (104 or xc2 i2 @xi( CM)]]} (lob) r Since vehicles j, Xi are involved in the accidents,with j play- ing the role of the initiator of the accidents, equations (lOa) and (lob) can be expressed as: (114 (lib) where i now represents the ith vehicle involved in the accident and j takes the value unity. The corresponding expressions to (11 a) and (11 b) in terms of the number of vehicles and the number of causes can be derived similarly from equation (6): d&f = f 6k,,lWW k,=l +c c ,f2{k~l-~ $ i 6kp,d1dt)} L ks=lp=l +r;2 {[ka?l ~k,,l@, cM) 1 (124 Appl. Math. Modelling, 1985, Vol. 9, August 277 A model of road accidents: J. 0. Asalor and E. A. Onibere and +; { f . . . fi. 1; 6kp,l(licM)) s=2 k,=l ks=l p=l K-I K-1 3 n(4,3) = w, 2) w, 3) c ~(~l~xl) I,=1 k,=l x,=2 K-l K-l I x Wl, x*1 + 1 c wl,xl)w,,xl) k,=l k,=l klfk, U5c) Other useful quantities which can be determined are: (1) Accident multiplier factor arising from interactions among vehicles and root primary causes H, defined as: f 4k 0) 2 40, r> k=l r=l HC (16a) 41, 1) = n(l , 1) (2) The ratio of all accidents to single vehicle accidents: Some potential applications Equations (11) express, for example, the total number of accidents in terms of the number of vehicles involved. Alternatively, they can be used to express the accident rates in terms of the rates of l-, 2-, . . . , r-vehicle accidents. Equa- tions (12) further express each accident class in terms of the causes, and particularly in terms of the number of causes that have interacted to cause the accidents. Since each vehicle that is involved in an accident must have a cause, according to the definition of the causation pro- cess,l the minimum number of causes that are involved in an r-vehicle accident is r. If an analysis of vehicles involved in accidents reveals that a proportion, n(k, i), are due to k causes and involve i vehicles, and ~(0, r) and n(k, 0) denote the total propor- tions due to r-vehicle accidents and k-causes respectively, then one has: H C n(k,O) k=l = H(l)+ l- s C n(k, 1) (16b) k=l This is an accident multiplier factor due to vehicular con- flicts. (3) The ratio of single vehicle accidents to single root primary cause accidents: H(l) + 1 E k=l 41, 1) (16~) where H(1) is the ratio of single vehicle accidents due to two or more r.p.c. to single vehicle accidents due to a single r.p.c. n(0, 1) = n(1, 1) + n(2, 1) + n(3, 1) + . . + n(S1, 1) n(O,2) = n(2,2) + n(3,2) + n(4,2) + . . . f n(Sz, 2) n(0, 3) = n(3,3) + n(4,3) + . . . + n(&, 3) n(0, r) = n(r, r) + n(r + 1, r) + . . . + n(S,, r) (13) where Si = i * S. Also: n(l,O) = n(1, 1) n(2,O)=n(2,1)+n(2,2) n(3,O) = iz(3, 1) + n(3,2) + n(3,3) n(k, 0) = n(k, 1) + n(k, 2) + . . . + ti(k, k) (14) The terms n(k, i) could be obtained from equation (12). So that n(k, 0) and n(0, r) could be evaluated. For example: K-l 1 @(k> 1) k=l 1 n(1, 1) = =- Qeff 4eff (Isa) K-l 42,2) = $W, 2) 1 Hk,, XI) (15b) (1) road condition, @( 1, 1) (2) vehicle condition, $(2, 1) (3) drivers safety ability, 45(3, 1) (4) pedestrian and other road users ability, @(4, 1) (5) circumstance, @(5, 1) k,=l wherej = 1. Computation of some of the quantities The special role played by circumstance needs to be further defmed to compute the above quantities from equation (12b). Circumstance plays no role in single vehicle accidents. Whenever two vehicles collide, they do so by virtue of circumstance featuring once, among other causal factors. In general, circumstance features exactly r - 1 times in an r-vehicle accident. Let the root causes be represented by: 278 Appl. Math. Modelling, 1985, Vol. 9, August A model of road accidents: J . 0. Asalor and E. A. Onibere Proportion of I-, 2- and S-vehicle accidents It is known that n(0, r) is the proportion of r-vehicle accidents. This is obtained from the summation of the terms n(k, i) as shown in equation (13). Mean number of causes per accident The mean number of causes per accident, MA, is given by: N S; (z, kg, kn(k, i) MA=---_ (17) where: $=S*i-i+ 1 n(k, i) = proportion of accidents involving i vehicles due to k causes excluding circumstance Discussion of results Using the variables shown in Table 1 and the constraint that @(l, 1) = $(2, l), the terms in sets of equations (13) and (14) were evaluated. From this evaluation, the ratio of I-, 2- and 3-vehicle accidents to total number of accidents was calculated with and without pedestrian influence. Also the mean number of causes per accident was computed. The Prime 750 computer of the Institute of Computer Science, University of Benin was used to perform the various computations. The calculations were based on the assumption that @(k, j) do not vary among the vehicles in the traffic system. This implies that all the drivers are exposed to an equal risk of accident per unit distance of road travelled. Figure 2 shows the proportion of all accidents involving 1, 2 and 3 vehicles for different values of circumstance. The results show that the proportion varies little irrespective of individual root primary causes other than circumstance. The proportion, however, is very sensitive to circumstance. The results also show that as the number of vehicles in- volved in an accident increases, the proportion of such accidents falls rapidly for accidents involving more than three vehicles and this is in agreement with references 9 and 10. In Figure 3, the mean number of causes per accident excluding circumstance is shown with or without pedestrian influence for extreme values of circumstance of 0.1 and 0.3. The results show that the mean number of causes per accident drops steadily as one of the factors other than circumstance becomes predominant. This agrees with expectation, since the predominant root primary cause requires less interaction in order to cause an accident. Also the mean number of causes per accident is sensitive to circumstance and increases with it rapidly. The limited evidence available agrees with the range of values shown in Figure 3. For example the average number of influencing factors per accident was 1.6 in West Germany and 1.5 in Switzerland in 1963.8 Conclusions A model of road traffic accidents that contains relatively few terms has been developed. The model is suitable for Circumstance @ (5.1) @(5,1)=0.1 1 vehicle 0 (5,l) = 0.3 1 vehicle a 0.2 / Ol- 0(5,1)=0.1 2 vehicles @(5.1)=01 3vehicles. I I I I I I I I 02 03 04 0.5 07 08 0.1 a Driver fact$% (3.1) : 1 o- 09- Circumstance @ (5.1) 5 o.a- @(5,1)=0.1 lvehicle ; 0.7- ; 2 0.6- 0 (5.1) = 0.3 1 vehicle 5 05- @ (5.1) =0 3 2 vehicles r g 0.4 e 03- a @(5.1)=0.3 3vehlcles 0.2. _ : _ _ _ _ . _ O.l- @(5.1)=0.1 2 vehicles/ 0 (5.1) ~0.1 3 vehGesl I I I I I I I I 02 03 04 0.5 0.6 0.7 0.8 0.9 b Driver factor @ (3.1) figure 2 Proportion of all accidents involving 1, 2 and 3 vehicles, (a) without influence of pedestrians, (b) with pedestrian factor o(4, 1) =0. 1 z 17- $! l.6- 0.3 2 1.2- 0.1 9 1.1 e 6 1.0 .+ I 2 I 1 I 1 I I I 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Driver factor o(3.1) Figure 3 Mean number of causes per accident (circumstance excluded) computer application in analysing road traffic accidents and their causes. The results obtained using the model agree with trends obtained from field investigations. Appl. Math. Modelling, 1985, Vol. 9, August 279 A model of road accidents: J. 0. Asalor and E. A. Onibere References 1 Asalor, J. 0. Appl. Math. Modelling 1984, 8, 133 2 Smith, A. M. and Watson, I. A. Reliability Engr. 1980, 1, 127 3 Fleming, K. N. and Hannaman, G. W. IEEE Transactions on Reliability 1976, R-25 (31, 171 4 Ganeloff. W. C. IEEE Transactions on Power Apparatus and Systems i975, PAS-94 (l), 21 5 Drew, D. R. Traffic flow theory and control, McGraw-Hill, New York, 1968, pp. 7-78 6 Wells, G. R. Traffic engineering, an introduction, Charles Griffin and Co., London, 1970, p. 133 7 Aaron, J. E. and Strasser, M. K. Driver and traffic safety education, Macmillan Publishing Co., New York, 1977, pp. 6-13 8 Babkov, V. F. Road conditions and traffic safety, Mir Pub- lishers, Moscow, 1975, pp. 10 and 20-21 9 Jones. I. S. The effect of vehicle characteristics on road acci- dents, Pergamon Press, Oxford, 1976, p. 18 10 U.S. Department of Transportation, National accidents sampling systems, 1982, DOT HS 806 530, March 1984, p. 28 280 Appl. Math. Modelling, 1985, Vol. 9, August