relatÓrio de trabalho no.06/04 -...

LEMT - Laboratório de Estudos Metodológicos em Tráfego e Transportes Departamento de Engenharia de Transportes - Escola Politécnica da Universidade de São Paulo

Caixa Postal 61549, CEP 05424-970, São Paulo/S.P., Brasil – (+55-11)-3818-5478, fax (+55-11)-3818-5716

RELATÓRIO DE TRABALHO No.06/04

PARAMETERS FOR EVALUATING PEDESTRIAN SAFETY PROBLEMS AT URBAN SIGNALIZED INTERSECTIONS

USING THE TRAFFIC CONFLICT ANALYSIS TECHNIQUE: A STUDY IN SÃO PAULO, BRAZIL.

Prof.Dr.HUGO PIETRANTONIO Departamento de Engenharia de Transportes - EPUSP

Eng.LUIZ FERNANDO BIZERRIL TOURINHO, M.Sc.

Departamento de Engenharia de Transportes - EPUSP

São Paulo, setembro/2004 (1ªedição) revisão 1 (setembro/2006)

LEMT - Laboratório de Estudos Metodológicos em Tráfego e Transportes Departamento de Engenharia de Transportes - Escola Politécnica da Universidade de São Paulo

Caixa Postal 61549, CEP 05424-970, São Paulo/S.P., Brasil – (+55-11)-3818-5478, fax (+55-11)-3818-5716

PARAMETERS FOR EVALUATING PEDESTRIAN SAFETY PROBLEMS AT URBAN SIGNALIZED INTERSECTIONS

USING THE TRAFFIC CONFLICT ANALYSIS TECHNIQUE: A STUDY IN SÃO PAULO, BRAZIL.

Eng.Hugo Pietrantonio, Doctor of Science in Transportation Engineering (email: [email protected]), Department of Transportation Engineering, Polytechnic School – University of São Paulo, Brazil.

Eng.Luiz Fernando Bizerril Tourinho, Master of Science in Transportation Engineering.(email:

[email protected]), Department of Transportation Engineering, University of São Paulo, Brazil)

Abstract: This paper reports on diagnostic parameters for pedestrian traffic safety problems using the Traffic Conflict Analysis Technique (TCT), particularly for pedestrian crossings at urban signalized intersections. The method of study is based on the U.S. FHWA-Federal Highway Administration guides for vehicular conflicts, applied using data collected from 26 pedestrian crossings of several types observed in 4 critical signalized intersections of the city of São Paulo, Brazil, all of them located on the Sao Paulo’s expanded CBD and experiencing high vehicular and pedestrians volumes. The adaptation of the technique for the observation and analysis of pedestrian-vehicle conflicts is discussed and related to previous works. The development of parameters has two tasks: the selection of an adequate classification of conflicts and/or segmentation of crossings and the determination of parameters for the classes (including recommended confidence levels). The results from field data are presented, including the count limits C (abnormally high level of counts) of pedestrian conflict that can be used in problem detection and the ratio R of accidents per million of conflicts (risk index of a type of conflict) that can be used in forecasting accident occurrences, for several alternative classification of conflicts and segmentation of crossings observed in the study. Usual statistical methods were applied with a new decision based criterion that could select the best classification/segmentation and the recommended confidence level to be used in applied work. Results obtained from the application of such methods to the case study are presented and discussed as an example application. . To our knowledge this is the first study offering these results for pedestrian-vehicle traffic conflicts and, despite limitations on the quality of our accident data, the parameters are consistent with the evidence presented in previous works.

LEMT - Laboratório de Estudos Metodológicos em Tráfego e Transportes

Parameters for Evaluating Pedestrian Safety Problems at Urban Signalized Intersections

1

1. INTRODUCTION This paper presents the results of a research aiming at establishing parameters for the diagnosis of pedestrian traffic safety problems in urban signalized intersections, distinguishing types of conflicts and types of crossings for signalized intersections, using the Traffic Conflict Analysis Technique (abbreviated here as TCT). Despite of failing to achieve widespread application, the need of evaluation methods based on direct observation of traffic operations, as the TCT, is a recurrent subject in academic and also in professional studies. Even after noting the effort in last years on developing practical safety performance models based on accidents, by themselves, a method of evaluation that use the direct observation of traffic operations is yet valuable for the safety analysis, by adding relevant information and responding in shorter time. As an advantage, the TCT has more abundant and confident data but this feature should be weighted against the risk of missing important factors in accident causation (for example, aberrant driver behavior and defects of vehicle parts will not reveal its role in TCT data). Due to a large international research effort carried-out during the eighties, the initial concept of traffic conflict was scientifically stated as “an event in which two road users (or a road user and another traffic element) became in a course of collision and an evasive action is observed (braking, swerving or accelerating) to avoid the potential accident” (see ICTCT/1984 for a historic background and the results of several studies of the international “calibration” effort). After the joint effort, leading institutions of several countries published guides for applying TCTs, including recommendations and criteria for its use in the diagnosis of road safety problems (e.g. HYDÉN/1987, MUHLRAD/1988, BAGULEY/1988, PARKER/ZEGEER/1989a, PARKER/ZEGEER/1989b). These publications are referred here as official guides (summarized in the appendices A to D), where one can distinguishes two ways in which the diagnosis using traffic conflict data can be carried-out. The basic diagnosis, reported in all TCT guides, is based on the interpretation of traffic count data by the professional analyst, relating them to traffic and site features and to the qualitative observations made during field work. The classification of traffic conflicts by type and severity aids the analysis but the task of identifying a set of relevant safety problems (and of selecting proposals for improvement of safety) remains an expert challenge, as in the process of diagnosis based on accident analysis. There is another type of diagnosis, recommended in the U.S.FHWA guide only (see PARKER/ZEGEER/1989a). The refined diagnosis, as we call it, tries to use objective parameters to carry-out two tasks:

- to identify types of traffic conflicts displaying abnormally high frequency compared to normal levels of traffic conflict counts for that type of traffic conflict (both referred to a standard 11 hour period of a week day) and

- to weight their accident proneness measured by the ratio of accidents to conflicts, or million of conflicts, counts (a measure of the risk level of each conflict and site type).

These refined diagnostic tasks, it should be stressed, are carried-out by the type of conflict considering the type of physical and traffic environment of the site and using objective diagnostic



2

parameters and can also forecast the expected accident frequency based on conflict counts, taking into account the conflict count for each conflict type and the type of site. One should also note that the viability of the refined diagnosis is constrained by the availability of previously calibrated parameters on the abnormal level of counts and on the accident risk ratio of traffic conflicts, by conflict type and type of site, developed in a careful and representative study of a set of similar sites. The U.S.FHWA guide was based on data from such kind of previous studies (GLAUZ/alli/1985, GLAUZ/MIGLETZ/1980), that distinguished some usual classes from unsignalized and signalized intersections, but there was little progress in developing new data since then. Despite this practical constraint, it is very important to stress the content of the refined diagnosis, underlining its meaning and relevance for the traffic safety analysis. For example, a validation study (PIETRANTONIO/1991, GUEDES/alli/1997) counted the same-direction and transversal conflicts between vehicles in an unsignalized intersection with medium traffic level:

- The expanded count for the standard period was 239,0 same-direction and 133,5 transversal traffic conflicts.

- Nevertheless, based on the U.S.FHWA data, the limits for abnormal frequency of conflicts are 410 and 24 (for a 90% confidence level), saying that the less frequent conflict type is the safety problem at the site.

- Also, again based on U.S.FHWA data, the accident/conflict ratio indicates that only transversal conflicts have a significant risk of generating accidents at this kind of site and, using the accident to conflicts ratio, it is possible to forecast a frequency of 13,6 collisions per year on weekdays with dry weather.

- The record in the previous year was 16 (all accidents), every of them being transversal collisions, as predicted.

This real case study clearly shows the importance of the diagnostic parameters in the analysis. In this setting, we built on our previous research and on the guidelines for TCT applied to vehicular conflicts (mainly the U.S.FHWA TCT) and searched for the development of diagnostic parameters for pedestrian-vehicle conflicts. In the following, we discuss the observation of pedestrian-vehicle conflicts, clarifying our methodological options, and review previous work on the typologies of conflicts and crossings to be used in studying pedestrian-vehicle conflicts in section 2. Section 3 presents the usual approach used for determining diagnostic parameters for the U.S.FHWA TCT and proposes a new decision based criterion that can be used for selecting the segmentation of crossings, classification of conflicts types and confidence level to be used in practical applications. Our study and the results we gathered are then presented and analyzed in section 4, with emphasis on the recommendations for practical application of the TCT on field studies. The final section summarizes the conclusions and underlines some suggestions for further research.



3

2. THE IDENTIFICATION, CLASSIFICATION AND ANALYSIS OF PEDESTRIAN-VEHICLE TRAFFIC CONFLICTS We identified five points that should be clarified for developing operational concepts of pedestrian-vehicle conflicts. Three general points are discussed, explicitly or not, in official guides (i.e. HYDÉN/1987, MUHLRAD/1988, BAGULEY/1988, PARKER/ZEGEER/1989a, PARKER/ZEGEER/1989b) and cover the delimitation of real conflicts (as opposed to “virtual” conflicts, taken as conflicts of slight severity or very small risk of generating accidents), the identification of their severity (of the conflict itself) and their level of risk (frequency and proneness of generating accidents). Two other points are recommendations driven to application, related to the typology of conflicts and the typology of sites that should be used for vehicle-pedestrian conflict studies, and were found only in some of the guides. Despite providing detailed instructions for field work and preparation of results in TCT studies of vehicular conflicts on intersections, the U.S.FHWA guide only sketch roughly the procedure for the observation of pedestrian-vehicle conflicts. A similar problem can be observed in TCT guides of other countries (e.g. HYDÉN/1987, MUHLRAD/1988, BAGULEY/1988). For vehicular conflicts, the U.S.FHWA guide provides the required diagnostic parameters for the usual conditions considered as well as parameters for sample design. The corresponding diagnostic parameters for pedestrian-vehicle conflicts are completely missing from the U.S.FHWA guide (and also from other guides). The recommendations for observation of pedestrian-vehicle conflict in several additional sources are too generic and their details are largely inconclusive. We classified these additional sources in two phases, considering their date relative to the intense work that resulted in the official guides (i.e. HYDÉN/1987, MUHLRAD/1988, BAGULEY/1988, PARKER/ZEGEER/1989a, PARKER/ZEGEER/1989b). The antecessor studies are mainly exploratory works (e.g. ZEGEER/alli/1980, CYNECHI/1980, ROBERTSON/1977a, ROBERTSON/1977b, U.S.DOT/1981) that used previous tentative concepts or classifications of traffic conflicts and other related events. The successor studies are predominantly applied works (e.g. GARDER/1989, JAVID/SENEVIRATNE/1991, CLARK/alli/1996, LORD/1994, ABDULSATTAR/alli/1996) and contain little advice on theoretical or operational concepts (but there are interesting hints in GARDER/1989, CLARK/alli/1996, ABDULSATTAR/alli/1996). In the following comments, we discuss the specific tips involved in observing and recording pedestrian-vehicle conflicts and describe our methodological options. 2.1. Delimitation of Relevant Traffic Conflicts (or, more generally, traffic events) Despite the clarity of the general concept, there are several practical questions that appear on the field observation of traffic conflicts and have to be judged. This point is related to the severity of



4

the traffic conflict (remember the Hydén prism of traffic events; see HYDÉN/1987, p.27) but also to the identification of the event of a traffic conflict and, perhaps, some other relevant events. The grading of severity will be discussed in the next topic. The other observational hints are covered here. In general, all guides achieve good agreement on the observational criteria that can be summarized as:

- Disregard preventive maneuvers (as the lowering of vehicle speed or the running of pedestrians without the presence of a course of collision with another conflicting user or element).

- Disregard virtual or potential conflicts (without significant risk of generating an accident, and that can be ignored).

- Devote special attention to quasi-accidents (conflicts with emergency evasive actions). - Include near misses (events with high risk of accident, given speeds and proximity, even

without a course of collision, perhaps by chance, but in which reaction time is very small and no evasive action could be taken if mandated).

- Distinguish conflicts from other traffic events, as traffic violations or user distraction (limiting its annotation to events that happen to generate traffic conflicts due to their occurrence). These events could be of interest also but then should be registered separately.

A point worth noting is that the U.S. guide is the only one that has a special concern with the identification of normal maneuvers that just give the right of way to users having priority on the road. These maneuvers are distinguished from conflicts (in which the evasive actions reacts to the danger of accident) and discarded (unless there is a clear sign of reaction to a potential accident, e.g. due to late perception of the user with right of way). The U.S. guide is also the only one that clearly identifies events with multiple conflicts from single conflicts and distinguishes primary from secondary conflicts (despite discarding secondary conflicts in the quantitative analysis of results). A multiple conflict forces several vehicles to take the evasive action (the number of reacting vehicles is the number of simultaneous conflicts). A secondary conflict is generated by the evasive actions of previous conflicts. The major simplification in the U.S. guide is the option of avoiding further classification of conflicts by severity level (after disregarding virtual conflicts and give-way maneuvers). Most of these points are not carefully discussed in all official guides and we felt it to be a relevant missing point, especially for the observation of pedestrian-vehicle conflicts. The analogy to vehicular conflicts can be supposed the hidden assumption in all official guides. The French guide (see MUHLRAD/1988, pp.49-53) includes a detailed description of pedestrian-vehicle (and motorcycle) conflicts, based on both vehicle and pedestrian evasive actions and accepts a clear analogy to vehicle conflicts. The Swedish technique even recommends the use of the same criteria (and values) for classifying the severity of pedestrian-vehicle and vehicular conflicts. These points will be discussed in the next topic also. The acceptance of the simple analogy of pedestrian-vehicle conflicts to vehicular conflicts can be disputed. On the field, the observation of who makes the evasive action is a clear distinguishing point related to pedestrian-vehicle conflicts, particularly when comparing the stopping (of pedestrians) and braking (of vehicles). The same point was discussed in other studies. For



5

example, Cynecki distinguished events with each type of actor doing the evasive action and decided to use only the observation on the driver maneuver for the determination of conflict severity, based on the same feeling (CYNICKI/1980, p.15). Finally, the event of a pedestrian hitting a vehicle is certainly unusual. The U.S. guide seems to extend this view to the identification of conflicts, counting only the events in which the vehicle is taking the evasive action (see PARKER/ZEGEER/1989a, p.17). One would count events in which pedestrians are taking the evasive action only if they are near misses. Another relevant point is that events involving multiple users are much more common when dealing with pedestrians (that usually walk in groups). The U.S.FHWA criterion for vehicle conflicts distinguishes the vehicle that generates the event and the one that takes the evasive action and recommends counting multiple conflicts based on the number of vehicles taking evasive action. If several vehicles are generating the conflict but only one takes an evasive action, it is counted as a single conflict. Note that secondary conflicts are more usual for vehicles and have the same nature (the involvement of more than one vehicle in the primary conflict), as vehicles flow in queues. Secondary conflicts are distinguished because other vehicles react to the second user of the primary conflict (if the other vehicles react to the same first user generating the conflict, then there is multiple conflicts). Multiple conflicts are discussed only by the U.S. guide and correspond to a rare event when observing vehicle conflicts. There is an option: count multiple conflicts as a separate type (what is similar to the criteria used for related accidents). Currently, when observing pedestrian-vehicle conflicts, the hidden assumption of analogy should be tempered using subjective criteria with more stringent requisites for discarding virtual conflicts (especially if the evasive action is taken by the pedestrian). The guiding idea is the same (to disregard events with ample reaction time) but the time scale is different. Thus, the idea of counting events in which the pedestrian does the evasive action only if the reaction time is very small can be recommended (it is a near miss if there is no course of collision and no evasive action). More stringent criteria should also be used for counting conflicts involving several users when multiple pedestrians are involved. They are counted as a single conflict, even when the group of pedestrians are taking the evasive action, if they are acting as one group (a mother and her children, a group walking “together”, …). When multiple vehicles or multiple groups take the evasive action, the number of vehicles or groups defines the conflict multiplicity. The alternative criterion of viewing single events with several users as one single conflict is a sensible option. The same can be said on using additional types of conflicts with multiple users. The concepts of primary and secondary conflicts are kept as usual. Following the U.S.FHWA recommendation, conflicts generated by other conflicts are counted as secondary conflicts and used only for the qualitative analysis (diagnostic parameters use primary conflicts only). Note that both criteria, for multiple and secondary conflicts, generate counts very similar to the alternative criteria that take them as single conflicts (perhaps it worth to note the added feature of involving other road users, that could be used in qualitative analyses).



6

2.2. Criteria for Identifying the Severity of Traffic Conflicts (and their use) Most of the discussion on severity in official guides is also limited to vehicular conflicts. The official guides, other than the U.S.FHWA one, give great attention to the identification of the conflict severity but its final importance to the objective analysis is small. In the established methods applied for vehicular conflicts, the severity is used only to identify and discard virtual conflicts (the lower severity level) and as qualitative information for the diagnosis. The Swedish TCT is an exception in which some studies related severity of conflicts to the risk of accident for vehicles and pedestrians (see HYDÉN/1987 and mainly ALMQVIST/HYDÉN/1994). The severity criterion of the Swedish TCT is based on comparing the value of the estimated TA variable against a critical value (TA is the estimate of Time to Accident and is defined as the time until the occurrence of the potential accident if the road users keep the same trajectories and speeds as practiced at the beginning of the evasive maneuver). In the application of the Swedish TCT, the TA is evaluated based on the subjective estimation of distance and speed carried-out by a trained observer. The calculation of TA and the classification of conflicts can be done in the field or at the office. In the current Swedish TCT, the same critical values of TA are used for vehicular conflicts and for pedestrian-vehicle conflicts (ALMQVIST/HYDÉN/1994). The critical values are derived from a vehicle braking curve (HYDÉN/1987, pp.117-118) and replaced the original constant value of 1.5 seconds (HYDÉN/1987). This criterion would be inadequate to grade conflicts in which the evasive action is not the vehicle braking. For example, in the practical application of Swedish TCT, we observed that the “official” criterion is often complemented by a subjective judgment when evaluating pedestrian-vehicle conflicts (see PIETRANTONIO/1999, in which a more general criteria, based on the available reaction time for the evasive action is suggested as a replacement, excluding the time of braking or other maneuver). The British TCT asks for the evaluation of four factors (time to accident, intensity and complexity of evasive maneuver, proximity of conflicting vehicles), judged subjectively by the trained observer. The classification of the severity is, then, done at the office based on a summary table for converting combinations of factor levels into a severity grade of a four level scale (BAGULEY/1988, p.33 e 34). The severity grade is initially used for discarding the slightest level of conflicts. The French TCT provides a careful description of a three level scale (light, moderate and severe conflicts) and asks for the subjective classification of road events by the trained observer (MUHLRAD/1988, item IV3.2). The classification has to be made by trained personnel during the field observation and, again, the slightest level of conflicts is discarded. Compared to the evaluation of the severity of each conflict in other guides, the U.S.FHWA TCT is really simplistic and only asks for identifying and discarding virtual conflicts (also described as conflicts with ample time for the evasive action), in the field work. Nevertheless, the U.S.FHWA guide is the only one that proposes a measure of degree of risk involved in the overall level of conflict frequency at a site (see the next item).



7

For pedestrian-vehicle traffic conflicts, the attention to the criteria for the grading of conflict severity decreased along the time. For pedestrian crossings, the conflict severity scale was evaluated in ZEGEER/alli/1980 (p.28) against constant values of TA (moderate severity between 1.0 and 1.5 seconds, severe under and slight over the moderate range). CYNECKI/1980 also evaluated the severity of pedestrian-vehicle conflicts but used a coarser scale (events with a distance greater than 7 meters between vehicles and pedestrians in the road, with or without evasive actions, were taken as slight conflicts; the events with a smaller distance and evasive action are classified as moderate conflicts and the quasi-accidents, i.e. those with emergency evasive action, are classified as severe conflicts). In the official guides, there is little advice other than the dubious suggestion of applying the same criteria as for vehicular conflicts to grade the severity of conflicts involving unprotected road users (as cyclists and pedestrians). Only the French TCT has specific recommendations based on its subjective grading procedure of a three level scale: slight (an unforeseen stop in the walkway or a simple accelerated walking), moderate (a sudden stop, jump back or running ahead when the vehicle brakes or proceed) and severe (a very rapid jump or a sudden jump when facing the vehicle body). Of course, the specific criterion is applicable only when the pedestrian takes the evasive action (the regular criterion should be applied when the vehicle does the evasive action). Based on our discussion, the subjective identification of virtual conflicts based on a subjective grading is recommended, as the one proposed in the French guide, but taking a more stringent criteria when the pedestrian takes the evasive action. As our practice was to use the grade of conflict severity only for discarding virtual conflicts, this seems to be enough and similar to other criteria, if not by noting that the evasive action is usually taken by the vehicle (we collected data for applying the Swedish severity grading, but the results were similar; see PIETRANTONIO/1999). 2.3. Criteria for Identifying the Measure of Risk in Traffic Conflicts The differences in the level of safety can be related with the relative frequency of conflicts and their probability of generating an accident (accident proneness) for each type of conflict and type of site (perhaps weighting the level of conflict severity and/or traffic conditions) as widely recognized in the official guides and other works. This is the content we attribute to the concept of level of risk for traffic conflicts (that we distinguish from the overall level of danger, a measure that should also weight the severity of the accident eventually generated in the events).

The naive view that raw numbers of conflict counts can be taken as measures of safety is clearly challenged in the U.S.FHWA guide, at least. Nevertheless, several works take the naive view yet. The first point to note is that different conflict types are intrinsically more common or rare in different types of site. This feature can be related to the process generating each conflict type in a type of site (distinguishing type of control and traffic variables). For example, a traffic signal usually separate



8

crossing movements and generate stop and go conditions in a giving approach. So, crossing conflicts become rare and same direction conflicts become frequent. One must consider the variation of the intrinsic frequency of each conflict type by type of site and analyze conflict counts in a relative way. This is the role attributed to the count limits C of normal traffic conflict level, differentiated by type of conflict and site. In the U.S.FHWA TCT, count limits are derived from the usual distribution of conflict counts for similar sites, both referred to a standard period, and a statistical confidence level has to be defined for getting the limit count C (PARKER/ZEGEER/1989a, provides C90% and C95% values). Of course, other relative measure could also be useful and a comparison of safe and unsafe sites could be a more robust approach to develop normal and abnormal conflict levels. Nevertheless, no other guide proposes a relative measure of intrinsic or normal conflict level. The second point to note is that the accident generation potential also varies by conflict type and type of site and must be distinguished. A clear measure, previously discussed, is the ratio R of accident to conflict, or million of conflicts, displayed by the U.S.FHWA guide for vehicular conflicts as a parameter to refine the diagnosis (weighting accident proneness) and to forecast the expected frequency of accidents, for each type of conflict in each type of intersection and level of traffic (signalized with medium and high traffic flow, and unsignalized, with low and medium traffic flow). This can be related to the degree of predictability and avoidabillity of an accident from a conflict type in a certain type of site. The Swedish TCT also determined accident to conflicts ratios (HYDÈN/1987, item 5.4 to 5.6), taking more general classes of conflicts and sites (signalized or unsignalized intersections, straight or turning movements, high or low speed sites, vehicles or “unprotected” road users). The French TCT proposed a risk matrix, based on the subjective evaluation of experts, to display the injury accident proneness in three levels (null/small, medium and high), with weights differentiated for signalized or unsignalized intersections and by type of conflict (MUHLRAD/1988, p.28-30), also considering conflicts with pedestrians and distinguishing some conflicts with motorcycles. Some studies evaluated the relationship between accidents and conflicts using other tools (e.g. JAVID/SENEVIRATNE/1991, LORD/1994, and also DAVIS/alli/1989). Nevertheless, the clear meaning of the ratio variables favors their use as a measure of risk. The option leaves only questions on the variability of these parameters for a chosen classification of conflicts and sites, that sets the level of detail needed to reach useful ratios. Note that there is no conceptual opposition between the risk levels defined in the French guide and the ratio of accidents to conflicts as the former can be taken as a clear proxy for the later. However, as the French guide does not distinguish the intrinsic frequency of different conflict types in different types of site, the meaning of the risk levels are unclear or mixed at best (perhaps including the potential of injury or death, as the French criteria is related to injury accidents). It seems that the same judgments can be extended to the study of the risk involved in pedestrian-vehicle conflicts but the calibration is more difficult as available data is scarce.



9

No data on the level of conflicts are available and the generic accident to conflict ratios developed by the Swedish guide are the only available information on accident proneness (HYDÉN/1987, p.71; see also ALMQVIST/HYDÉN/1994 in which ratios vary by severity level). Table 1 has a sample of ratios that will be used for comparison with our empiric estimates, reported ahead.

Table 1 – Ratio of Accidents per Million Conflicts from Swedish Studies. Sweden/98

– All Severe Bolívia/94 – Low Severity

Bolívia/94 – High Severity

Car-Car “parallel” 28 10 60 Car-Car “right-angle” 119 40 200 Car-Unprotected Road User 339 200 700 Source: ALMQIVST/1998, PIETRANTONIO/1999, ALMQVIST/HYDÉN/1994 (transformed to Accidents per Million Conflicts) The risk levels attached to pedestrian-vehicle conflicts in the French guide are the other available (but coarse) data. For the proneness of a moderate or severe conflict generating an injury accident, both for unsignalized and signalized intersections, the risk is set as small/null for parallel and right turn pedestrian-vehicle conflicts and medium for straight and left turn pedestrian-vehicle conflicts. The scarcity of data for analysis is noticeable as the classification (or segmentation) problem has the drawback of requiring the availability of large samples for significant statistical results. In our setting, good conflict and accident data are needed (what is even worse when dealing with pedestrian-vehicle conflicts and accidents).

So, it seems unavoidable that knowledge in this subject should progress through a series of individual studies, devoted to specific samples.

2.4. Typology of Pedestrian-Vehicle Conflicts There is no general agreement on the more convenient typology of pedestrian-vehicle conflicts, but there is a clear trend to simplified categories based on the type of movements for each user (as in the typology of vehicle conflicts). A fundamental reason behind this option is the desire of adopting a typology similar to that conventionally used for accidents. Both official guides that treat this question (MUHLRAD/1988, PARKER/ZEGEER/1989a) recommend 4 types of pedestrian-vehicle conflicts similar to vehicle conflict types. These can be combined in 5 types (the U.S.FHWA separate the conflicts with straight vehicles based on their position before or after the intersection, while the French guide aggregate them but includes a category of conflicts with pedestrians and vehicles in parallel movements). The British guide does not deal with pedestrian-vehicle conflicts and the Swedish guide does not recommend a standard typology. Former studies (e.g. ZEGEER/alli/1980, CYNECKI/1980, U.S.DOT/1981) distinguished several types of events based on the way the pedestrian approaches the road, on the vehicle movements



10

involved and on the type of traffic violation observed (13, 12 and 12 types in ZEGEER/alli/1980, CYNECKI/1980 and U.S.DOT/1981, respectively). More recent studies have kept a smaller level of detail. For example, CLARK/alli/1996 (p.41) used 5 types, classified based on type of evasive action (that would be difficult to use in a disaggregate analysis, despite being elucidative in the study). Some studies even treated conflicts aggregately. However, the studies using aggregate data faced significant problems (particularly for relating conflicts and accidents as in DAVIS/alli/1989 and SAYED/ZEIN/1999), what is evidence on the importance of an adequate typology of pedestrian conflicts also. Our previous research (PIETRANTONIO/1999) used 8 types of pedestrian-vehicle conflict based on the U.S.FHWA guide (2 other types would be added based on the French guide), distinguishing the pedestrian direction of flow relative to the vehicle (as in Figure 1), and recorded conflicts by pedestrian crossing (instead of vehicle approach). Nevertheless, our results were unable to demonstrate the relevance of the added detail (direction of pedestrian flow).

1 P/TPd 2 P/TPe 3 P/TAd 4 P/TAe

1 P/TPd: pedestrian from right of straight vehicle, near crossing 2 P/TPe: pedestrian from left of straight vehicle, near crossing

3 P/TAd: pedestrian from right of straight vehicle, far crossing 4 P/TAe: pedestrian from left of straight vehicle, far crossing

5 P/TDF 6 P/TDR 7 P/TEF 8 P/TER

5 P/TDF: pedestrian to frontal path from right turning vehicle 6 P/TDR: pedestrian to back path from right turning vehicle

7 P/TEF: pedestrian to frontal path from left turning vehicle 8 P/TER: pedestrian to back path from left turning vehicle

Figure 1 – Pedestrian-Vehicle Conflict Types in Intersection Crossings

Our study also show the difficulty involved in using the same typology for accidents, giving the lack of clear information for recovering the exact crossing location and pedestrian/vehicle movements in police accident records.



11

As the pool of evidences is again scarce, we feel that additional study should be devoted to this question. The use of similar typologies for conflicts and accidents is a practical advantage and the need of more detailed classes is a research theme. Currently, we keep on using the 4 classes typology based on the U.S.FHWA guide (as the parallel movements conflict is rarely observed and identified in intersections). However, from our experience, the annotation of details on the events is recommended, useful also as a way for studying and developing alternative typologies. The use of the Swedish record sheet (see ALMQVIST/1998, ALMQVIST/HYDÉN/1994) or the annotation procedure of recording one conflict in each line of sheet suggested by Hummer (ROBERTSON/alli/1994) for the U.S.TCT (and appending relevant comments on the events), both would favor this task. 2.5. Typology of Pedestrian Crossings in Signalized Intersections The U.S.FHWA guide suggests the use of the overall intersection, instead of each approach as the unit of analysis in the study of vehicle conflicts, at least for the refined diagnosis. One can question this option for the study of vehicular conflicts and, even more, for the study of pedestrian-vehicle conflicts. Of course, using more detailed data could bring a higher coefficient of variance and then would request more hours of observation to reach similar statistical quality. Nevertheless, the aggregate unit can lose significant detail and would multiply the number of possible types of sites, perhaps reaching a prohibitive level when considering pedestrians. So, we treated crossings individually. With the intersection as its unit of analysis for the refined diagnosis, the U.S.FHWA had suggestions of parameters for four types of intersections: high and medium flow signalized intersections and medium and low flow unsignalized intersections, all these cases considering four leg junctions of two way approaches (some other studies analyzed three leg junctions). At the technician risk, one can apply the provided parameters for intersections with peculiar features (some one-way approaches or other distinctive feature) with added care in the analysis of results (as in the successful real case that was commented on the introduction of this report). The other guides had no suggestion for typology of sites. However, when studying accident to conflicts ratios, the French guide differentiated signalized and unsignalized intersections in the risk matrix and the Swedish guide further segmented unsignalized intersections as low and high speed in developing accident to conflict ratios. Official guides have no specific recommendation for the analysis of pedestrian-vehicle conflicts, at the intersection or crossing level. Among former studies, CINECKY/1980 (p.12-13) made an exploratory discussion of relevant features, taking a large number of variables.



12

Among more recent studies, GARDER/1989 (p.440) selected a typology similar to the one used by HYDÉN/1987, with three classes: signalized intersections, low speed or high speed unsignalized intersections (the effect of several other variables were also studied). Studies about accidents (e.g. ZEGEER/1991, ZEGEER/alli/1982) also showed a similar pattern of classes and variables. As the typology of sites is used for establishing diagnostic parameters (and then determine the level of effort needed for collecting field data), the segmentation should be revealing and parsimonious, at the same time. We were not able to find any conceptual discussion of the criteria for classifying sites and adopted the distinction between essential features (that change the way road users interact) and residual features (that vary the level of safety in the interaction), on the hypothesis that the first group of features must always be distinguished in the classification analysis. Admitting also that the crossing is a better unit of diagnosis, our previous research on signalized intersections with medians or one way streets (PIETRANTONIO/1999) segmented pedestrian crossings in two main groups: entry (near or stop) line crossings and exit (far or free) crossings. Of course, crossings of two-way approaches without median crossings form a complementary type and similar types on unsignalized intersections are also natural candidates. Criteria for further segmenting pedestrian crossings, even for signalized intersections with medians, are not evident. The existence/absence of vehicle turning movements and of opportunities for red-running pedestrian crossings (that can be related to the availability of gaps on the vehicle stream) should be studied. Ranges of flow or speed are also natural criteria (as suggested by other studies or other settings) as a proxy for exposure measures (i.e. relevant at least for count limits, if not for risk ratios also). The type of treatment for pedestrians in signal phases (as exclusive/protected or concurrent/permitted pedestrian phases) seems to be next candidate as essential feature. Other factors, as the existence of pedestrian signal heads, of a painted or signed pedestrian crossing, the use of lay-outs with pedestrian refuges or displaced crossings, the use of signal plans with one stage or two stage crossings, there are all candidates even if less probable. Anyway, larger samples will be required to develop parameters for finer crossing categories.



13

3. METHODS FOR DETERMINING AND EVALUATING DIAGNOSTIC PARAMETERS FOR PEDESTRIAN-VEHICLE CONFLICT STUDIES The diagnostic parameters used in the refined procedure associated to the U.S.FHWA TCT are based on observations of pedestrian-vehicle conflicts by type of conflict and type of crossing. The development of diagnostic parameters has two tasks: the selection of an adequate classification of conflicts and/or segmentation of crossings and for the estimation of parameter values for the selected classes (including confidence levels when required). The procedure for determining the limit counts C for traffic conflicts recommended in the U.S.FHWA guide (PARKER/ZEGEER/1989a, see also GLAUZ/MIGLETZ/1980) can be applied without major difficulties, after gathering the required data (conflict counts, expanded for the standard period). In this method, the abnormal level of the frequency of conflicts is determined through the analysis of the distribution of daily conflict counts per standard period, for each type of conflict, on similar sites of a class (procedures for collecting conflict counts and expanding the data to a standard period are simple and are also suggested in PARKER/ZEGEER/1989a, the U.S.FHWA guide). Counts are usually obtained from short period observation (around 30 minutes each) and expanded for the standard period, filling the periods without counts based on an average rate (for workdays, the standard period of eleven hours recommended runs from 07 a.m. to 06 p.m.). The procedure for determining the ratio R of accidents per conflict, or million of conflicts, is not described in the U.S.FHWA guide or related works (i.e. PARKER/ZEGEER/1989a, GLAUZ/alli/1985, GLAUZ/MIGLETZ/1980). Nevertheless, the diagnostic parameter is a ratio of two random variables and available methods for ratio estimators are widely discussed in statistical textbooks on sampling theory (e.g. COCHRAN/1977). For achieving this task, traffic accident data has to be gathered and referred to the same observational units (intersections, approaches or crossings) and traffic conflict data has to be expanded to the same time frame of reference (e.g. workdays of a year). The procedures for expanding conflict counts for a year are presented in the U.S.FHWA guide (PARKER/ZEGEER/1989a, in chapter 8, when discussing the forecasting of accident frequency using conflict data) can be used in this task of converting daily counts to yearly totals. The evaluation of the diagnostic parameters should also be based on statistical criteria. There are conventional and simple recommendations on this subject but a decision criterion that is more adequate for engineering work seems to be preferable and is used in this work. For completeness, the estimation procedures are shortly described in the following topic. 3.1. Statistical Methods for Determining Diagnostic Parameters for TCTs The determination of diagnostic parameters for TCT has two peculiar features: - basic variables have non-gaussian distributions and - the risk measure is a ratio of random variables. For establishing count limits C, raw data are the expanded daily conflict counts for a standard period (e.g. the 11 hour period ranging from 7 a.m. to 6 p.m, for workdays, suggested in the U.S.FHWA



14

guides). Accepting the empirical evidence that expanded daily conflict counts have a Gamma distribution (HAUER/1975 apud GLAUZ/MIGLETZ/1980), their statistical parameters (s,t) can be estimated using some statistical method. For any statistical confidence level %L , count limits %LC can be obtained accordingly using the parameters of the fitted distribution for similar sites and each type of conflict, as sketched in Figure 2.

Figure 2 – Gamma Distribution and the Limit Counts for 90o and 95o quantiles. (based on [5])

Following the U.S.FHWA guide (see PARKER/ZEGEER/1989a and also GLAUZ/alli/1985, GLAUZ/MIGLETZ/1980), given a sample of sites, the estimates of the parameters using the sampling method of moments are

2C

Cs

mt = (1)

Cm.ts = (2) with the average Cm and the variance 2

Cs of daily counts, expanded to the standard period. The count limits %LC for similar sites and each type of conflict, for selected statistical confidence levels %L , can be determined using Chi-square tables and the relationship between Chi-square and Gamma random variables (see PARKER/ZEGEER/1989a, pp.62-63). Knowing that the critical

value of the Gamma distribution (the count limit) is given by t.2

XC

2%L,

%Lν= , where 2

%L,Xν is the

critical value of the Chi-square distribution with s.2=ν degrees of freedom corresponding to the same L% confidence level (or percentile), calculations can be carried-out with a standard Chi-square table, for the selected values of %L . Given the parameters of the fitted distribution and applying the formulas above, one has to calculate ν and interpolate for the value of 2

%L,Xν in a Chi-square

table, before the count limits %LC are calculated.



15

Nowadays, all calculations can be carried-out directly with simple spreadsheet software (as example, MS-Excel has the functions GAMMADIST and GAMMAINV for the density/cumulative gamma distribution and its inverse cumulative probability function, using the more usual description of the Gamma distribution with parameters sa = and t

1b = ).

There is no suggested procedure for the statistical evaluation of the estimated count limits or their performance in identifying unsafe sites in the U.S.FHWA guide (PARKER/ZEGEER/1989a). A conventional option could evaluate the fitting of the statistical distribution with the Kolmogorov-Smirnoff goodness of fit measure D. Using the previously quoted statistical properties and being i the index of each count iC in increasing order of count value, LAW/KELTON/1991 (pp.387) points that the sample value of the Kolmogorov-Smirnoff statistic can be evaluated as

−−−=

niFF

niD

ii CCi1,max , where

iCF is the cumulative probability at i

2i,v C.t.2X = for

s.2=ν in the standard Chi-square table (then standard statistical tests can be applied). Nevertheless, a decision based criterion for analyzing the performance of the count limits, selecting the best classification of conflict types and/or segmentation of sites and also the best confidence level for practical applications is suggested in the section 3.2 ahead. For establishing the risk measure R, and its variance, several conventional methods of determining ratio estimators of sampling studies can be used. The most common estimators are the ratio of means (or totals) and the mean of ratios. Both are biased estimators and COCHRAN/1977 (see chapter 2 or 6) suggests the use of the ratio of means (or totals) as it delivers a smaller mean square error (where E[R]²=Bias[R]²+Var[R], with Bias[R] the expected bias of the estimator and Var[R] its variance). The analysis considers all sites of the same segment or a pool of segments, for each type of conflict or some aggregate type. For a class, with the total number of accidents ( At ) and conflicts ( Ct ) for all the sites in the sample or the average values of accidents ( Am ) and conflicts ( Cm ) in the sample of sites, the estimators of the ratio R and its standard deviation are:

C

AEi

iEC

A

mm

CA

tt

R̂ ===∑∑ (3)

( ) ( ) ( )1n

C.R̂C.A.R̂.2A.

n.m1)R̂(s

2Ei

2Eii

2i

C −+−

= ∑∑ ∑ (4)

where iA and E

iC are the expanded frequencies of accidents and conflicts for each site and n is size of the sample of sites for a given class (see COCHRAN/1977, pp.30-34). The expansion of conflict data to totals per year are carried-out using notional conversion factors (instead of using methods based on statistical distributions or empirical coefficients) as in the procedure recommended by the U.S.FHWA guide (PARKER/ZEGEER/1989a). The advantage of using notional conversion factors is that they do not contribute to inflating parameter variance. The disadvantage is a potential reduction in the precision of the risk measure.



16

However, note that the decision of using the same time period for accident and conflict data is not mandatory as long as they refer to the same type of period. The option of relating data obtained from the same time frame but using different unit of time is a viable one (that would avoid complex extrapolations). These parameters are called rough ratios (RR, e.g. in yearly accidents per hourly conflicts) and are best suited for practical purposes (even if less precise). For both ratios, the statistical quality of the estimate is measured by its standard deviation, its variance or, relatively, by its coefficient of variation (the inverse of a “t” statistic). Tests on the difference between ratios, with the assumption of independent samples, can be carried-out only approximately (as the distribution of the ratio estimators is non-normal and unknown), based on a standard t test of the quasi-t statistic (see COCHRAN/1977, pp.180-183). Then, the comparison of the difference of ratios on two disjoint segments of crossings can be carried-out (at least approximately) and its statistical significance asserted as long as the factors that generate dependence between counts on the sites of each segment are of minor importance. The practical performance of the ratios as estimators in forecasting accident frequency can also be analyzed by evaluating the expected error of estimates, following the procedures recommended in the U.S.FHWA guide (PARKER/ZEGEER/1989a). However, the comparison of predicted and observed accident counts (or yearly frequencies) on a sample of sites is also possible, as long as the sample is not subjected to usual selectivity biases or the regression to the mean effects is accounted for. This option is more revealing by displaying the explained and unexplained variation of the conflict ratio forecasting method and was carried-out in the application. An external validation study (with other sample) would do better, evaluating transferability also. The option of comparing the forecasting errors achieved (expected or observed) can evaluate several different methods (e.g. accident to conflict ratio, accident to exposure ratio, accident regression models, empirical Bayesian procedures). Nevertheless, it is important to weight the content of each method and the way it is applied in studies. For example, the TCT method is almost unique in using observational tips from the traffic operation (a competitor would use, at most, basic variables as traffic volume or speed) and can be applied immediately (so does not have to wait for accidents to happen). The error is just one of the things that matter. 3.2. A Method for Evaluating Segmentations and Confidence Levels for Diagnostic Parameters of TCTs with a Decision based Criterion Instead of searching for purely statistical criteria, the performance of alternative segmentation or confidence levels can be evaluated using a loss function based on the benefits and costs of implied decisions that will be motivated by the use of the parameters. The same approach can be applied to the other settings as well, with small methodological changes. For TCTs, these decisions correspond to the selection of sites for treatment. The costs and benefits are related to:

- the realization of the potential accident reduction with treatment of risky sites; - the waste of resources devoted to safe sites also selected for treatment; - the saving of resources of avoiding the treatment of safe sites; - the remaining social costs of accidents for risky sites not selected for treatment.



17

Each effect must be evaluated given parameters used in the diagnostic activity that select sites for treatment and data available on the partial effectiveness of treatments (taking into consideration the persistence of unavoidable accidents, at least with usual countermeasures). An internal validation study can be carried-out applying this analysis to the sample used for the establishment of parameters. An external (or cross) validation study would conduct this exercise in another sample of sites and will evaluate the transferability of the parameters as well. The evaluation can be extended to the monitoring of costs and benefits of the treatments or can use average or representative measures of potential costs and benefits. Using this terminology, the application that will be reported in this study will do an internal validation with representative measures of costs and benefits. Our procedure supposes that the TCT is used in selecting sites for treatment by comparing average daily counts (for any number of days) to the count limits of a given confidence level, then deciding to treat the sites with abnormal frequency of conflicts of any type. The limit counts can be developed with alternative segmentation proposals (even aggregate). The procedure also supposes that all selected sites are treated and that representative values of costs and benefits are available for the four possible cases, as depicted in Table 2.

Table 2 – Approach for the Validation with Economic-based Weights and Representative Values.

Size of the Validation Sample (N)

Sites with record of Accidents (NB)

Sites without record of Accidents (NN)

Normal sites (NS) with conflict counts under Count Limits.

Error Type I: NEI (Neglected Risk)

Costs of Avoidable Accidents

HIT Type II: NHII (Saving of Resources)

None (or benefit of alternative use) Abnormal sites (NA) with conflict counts over Count Limits

HIT Type I: NHI (Detected Risk)

Benefit of Safety Improvement

Error Type II: NEII (Wasting of Resources)

Cost of Intervention

• Type I Error Neglected risk: not selecting a site that has accident records in the sample information. The economic weight is related to the costs of avoidable accidents (e.g. 50% of US$ 20.000/acc for direct costs of accidents).

• Type II Error Wasting of resources: spending money for treating a site without accident record in the sample information. The economic weight is related to the cost of usual treatments (e.g. US$ 5.000/crossing for small interventions).

• Type I Hit Detected risk: selecting a site that has accident records in the sample information. The economic weight is related to the reduction of accidents less treatment cost (e.g. 50% of US$ 20.000/acc less US$ 5.000/crossing)

• Type II Hit Saving of Resources: avoid spending money for sites without accident records in the sample information. The economic weight is null (or the average benefit of alternative uses of resources in other areas can be used)

We prefer to communicate benefits and costs in US$ (generating a kind of economic weighted index) but, of course, any other agreed compensatory scale can be used (even non-monetary). The values shown are representative of direct or material costs of accidents with average severity in Brazil and were selected for illustrative purposes on conservative grounds (other criteria, including human capital and/or pain and suffering components, could be also added as well). With economic based weights, the overall evaluation criterion is a kind of net economic index, following usual criteria applied in Benefit/Cost Analysis. Using the hypothetical representative weights suggested in Table 2, the net economic index E can be calculated with data on the number



18

of sites for each represented cell (NEI, NEII, NHI, NHII) and the corresponding number of accidents (AEI, AHI only on error type I and hit type I sites). The number of sites with accident is

NEINHINA += , the number of sites without accidents is NEIINHIINWA += , the number of treated sites is NEIINHINTS += , the number of avoided accidents is AHIAA = and the number of remaining accidents is AEIAR = . Admitting that accidents are only partially avoidable through engineering treatments, the costs of unavoidable accidents can be ignored. Nevertheless, the treatment costs must be computed on all sites selected for treatment. A restricted net economic index would evaluate benefits at treatment sites only (ignoring the remaining accidents on non-treated sites with accident records) but a better measure of net economic performance can be constructed noting that the “do nothing” scenario has a large negative net economic index that should be corrected. With weights given by Sα (potential accident savings) and Cβ (treatment costs) as defined in Table 2, the net economic index of avoidable costs in the “do nothing” (E0) option is

( )AEIAHI.A.0E SS +α−=α−= (5) and the net economic index attained with treatments selected based on a given criterion is

( ) AEI.NEIINHI.AR.NTS.E SCSC α−+β−=α−β−= (6) (that can be better than the “do nothing” measure even if negative). Based on conventional Benefit/Cost Analysis, a differential measure of the net economic index against avoidable costs in the “do nothing” scenario (DE) should be built as

NEII.NHI.AHI.0EEDE CCS β−β−α=−= (7) that is the usual restricted measure of economic performance based on treated sites only (as the number of avoidable accidents on non-treated sites cancel out). The maximum attainable net economic index of treatment options (ME) is

( ) )NEINHI.(AEIAHI.NA.A.ME CSCS −β−+α=β−α= (8) (delivering the measure of performance for an ideal criterion) and a relative measure of performance over the maximum attainable (RE) is

( ) ( )NEINHI.AEIAHI.NEII.NHI.AHI.

MEDERE

CS

CCS

+β−+αβ−β−α

== (9)

that is the improved economic measure previously quoted (considering all observed accidents in the validation sample). The best segmentation and confidence level for decision is the one that delivers the greater value of the performance measure, provided that its net economic index is better than the “do nothing”



19

option. Even if not a perfect option (approaching 100% of ME) or if not the best possible option, any criterion with E>E0 (or DE>0) is defendable. The worst possible result is NWA.WE Cβ−= but any policy delivering DE<0 is worse than the “do nothing”. The performance measures need not deliver a positive net economic index per se (as the status quo carries significant social costs) but the treatments should attain usual positive net economic return, represented by the differential net economic index DE (preferably with a large ME). These set of measures also have other very convenient properties:

- The differential measure DE, being a linear combination of results on the number of sites and/or accidents in each outcome cell, delivers the overall achievement of any segmentation as the summation of the DE on each segment, revealing where the best or worst contributions to the overall results were gathered. The segments where the DE partial achievements have poor performance are natural candidates for redefinition or further segmentation.

- The relative measure RE, being bounded by 100%, has a clear meaning for measuring the final performance against a tight limit. Any criterion with positive DE is usable, based on economic ground, but criteria that can reach RE measures of 80% or 90% are also at the limit of the possibilities for whatever criteria (at least for those calculated based on the same set of established parameter values).

This evaluation approach can not be used directly to the validation of the ratio R of accidents to conflicts as there is no clear decision related to its use in the conventional approach. The previously quoted options of evaluating the statistical significance of differences in R for alternative classification/segmentation (using the quasi-t statistic with the standard t test as an approximation) and of evaluating the accident estimates obtained with the accident to conflicts ration against observed accident frequencies, both were used in the analysis of parameters values. Nevertheless, in the comparison with observed accident frequencies, the technical requirement of using individual pedestrian crossings as units of analysis (instead of intersections) is a problem, given the small number of accidents in each observation. Before leaving the theoretical discussion, two additional comments are worth of notice. First, note that a decision criterion to select sites for treatment based on the accident estimates obtained from conflict counts can be proposed (being practical or artificial, i.e. that will be used for evaluation purposes only). Then, ratios would be amenable to the same decision based method of evaluation applied for the classification/segmentation and confidence level of count limits. The criterion could be based on the expected number of accidents predicted (e.g. greater than 1 in a year) or on some figure derived from an approximate distribution of the estimate (e.g. more than 75% chance of a pedestrian accident in a year). In each case, the selection of value for the criterion could be also carried-out with the proposed method. Second, note that the proposed method is reminiscent on the view of traffic conflicts as proxy to traffic accidents (taken as the “true” measure of unsafety, and as the only one that should be considered and whose reduction would be weighted as benefit of the actions). A wider view considers a traffic conflict as a safety measure per se and as an event that brings discomfort to road users even if the danger of accident is avoided (as any other related traffic event that burdens on safety). Then, one can estimate the value to road users of reducing traffic conflicts (i.e. the “cost” of traffic conflicts, similarly to the “cost” of accidents, travel time and the like). Its inclusion in the net



20

economic indexes is fully justified based on a Benefit/Cost rationale, at least as long as it represents objective discomfort. However, as a measure of accident risk, its role is played in the forecasting of accident frequencies (and double counting should, then, be avoided). This same reasoning would suggest that other effects of safety measures (as the increase in travel time brought by a reduction of speed mandated by safety goals) should also be included as well. These points are limitations that could ask for improvements when evaluating other techniques or some specific countermeasures. In the case study, representative values were used and a sensitivity analysis carried-out for most parameters values, as reported in the following section, but always keeping the initial approach described here of weighting only the benefits of accident reduction.



21

4. APPLICATION TO THE DETERMINATION OF DIAGNOSTIC PARAMETERS FOR PEDESTRIAN-VEHICLE CONFLICTS IN THE SÃO PAULO STUDY The diagnostic parameters for pedestrian-vehicle conflicts were obtained based on the 1998 study carried-out in 26 pedestrian crossings of four critical signalized intersections of the City of São Paulo (see PIETRANTONIO/1999). The main interest here is on the description of the application of the decision based criterion and the way it fills the role of supporting the development of parameters, mainly the selection of segmentation and confidence level. The study will be summarized for the interested reader in section 4.1 (that can be skipped without loss of continuity). The main analysis is carried-out in section 4.2 that also present the main results. Conflict types were based on the preliminary classification shown in Figure 1, previously. Table 3 has a simplified sketch of each intersection and shows the position of each pedestrian crossing (using its number in the intersection). The abbreviated notation used for identifying crossings is cumbersome but is also kept here for precision. The intersections are identified in the corresponding column of Table 3. Two intersections are located in the old central area (Ip-SJ and Co-CP). The other two are located in southwest of the expanded central area (FL-TS and FM-VB). Pedestrian crossings were classified and nominated as done in the original study (PIETRANTONIO/1999), as identified in the corresponding column of Table 3. Common features of all signalized intersections and their crossings are:

- high vehicular and pedestrian flows, with averages around 1.500 vehicles and 1.000 pedestrians per hour in each crossing;

- road markings (mainly painted crossings and stop lines) in good conditions, where provided, but significant pedestrian movements were observed also in some unmarked crossings;

- the flows related to commercial activities and transit services is very important in all the areas;

- carriageways have at least two lanes and are one-way (two-way roads have separated carriageways with raised medians);

- there are signal groups for vehicle and pedestrians at all intersections but one (the FL-TS intersection);

- especial lanes and especial phases for turning vehicles are absent and left turn are locally forbidden and rerouted through loops on adjacent streets.

Vehicle and pedestrian flows are also shown in Table 3. Traffic operations at the sites were registered with video-cameras in the morning, mid-day and afternoon peaks (2 hours in each period), along with the counting of conflicts by field personnel. Due to a failure in the positioning of the video-camera at one site, recording of pedestrian flows were missed on one of its crossings and the corresponding data had to be discarded from the analysis.



22

Table 3 – Basic Data on the Intersections and Crossings of the 1998 Study. a. General Data on the Intersections of the Studied Pedestrian Crossings.

Intersection Identification of Roads No.Approaches No.Crossings Obs.:

Ip-SJ Ipiranga Av.

X São João Av. 2 x 1 6 Both approaches of Ipiranga Av. have

the same direction of flow.

Co-CP Consolação Av. X Caio Prado St. 2 x 1 7

Caio Prado St. split into two exiting roads after the intersection

FL-TS Faria Lima Av.

X Teodoro Sampaio St. 2 x 1 7 Major right turn from the Teodoro

Sampaio St. before the intersection

FM-VB Francisco Morato Av.

X Vital Brasil Av. 3 x 1 6 One-way busway to the CBD at the center of the Francisco Morato Av.

b. Data on the Studied Pedestrian Crossings of Each Intersection.

Intersection Crossing Hourly

Vehicles Hourly

Pedestrians 96/97 Accident Location

Index (%) Intersection Sketches

and Sites of Crossings

1: Sj-TP 1021.20 2073.60

2: SJ-TA 1299.20 1932.20 1-100% (AC)

Ip-SJ 3: Ip-E-TP 1165.20 1287.20 1-60% e 100%(AC)

4: Ip-D-TP 2197.40 1287.20 1-40% (AC)

5: Ip-E-TA 1475.40 1125.60

6: Ip-D-TA 1609.20 1125.60

1: Co-TP-BC 4146.80 1007.60 1-100% (AC)

2: Co-TA-BC 4146.80 -

3:Co-TP-CB 4463.40 -

Co-CP 4: Co-TA-CB 3931.60 1088.00

5: CP-TP 978.00 654.80

6: CM-TA 488.80 747.40

7: MA-TA 1166.80 682.20 2-100% (AC)

1: FL-TP-IP 1312.40 756.80 2-60% (AC)

2: FL-TA-IP 1347.40 902.40 2-40% (AC)

3: FL-TP-PI 1096.60 902.40 1-67% (AC)

FL-TS 4: FL-TA-PI 1096.60 756.80 1-33% (AC)

5: TS-E2-TP 1343.80 556.40

6: TS-E3-TA 1343.80 55.60

7: TS-TA 1290.80 849.60

1: FM-TP-BC 1700.00 1215.60 1-75%, 1-60% e 1-100% (AC)

2: FM-B-TP-BC 375.60 1415.10* 1-25% e 1-40% (AC)

FM-VB 3: FM-TA-CB 1334.10 1142.88

4:VB-D-TP-BC * *

5:VB-TP-BC 867.50 1186.32

6: VB-TA-CB 706.12 1142.88 1-100% (AC) Average Hourly Flow 1557.17 1044.26



23

Finally, accident data from police records are summarized in Table 4. Note that only part of the registered accidents had the official police accident reports recovered.

Table 4 – Summary of Pedestrian Accidents and Recovered Police Reports in the Intersections.

Pedestrian Accidents – 1996/1997 Intersection Year Total

Ped.Acc. Work Days

Week Ends

Recovered in each year

Total and % Recovered

in workday standard period

96 8 7 1 3 1

Ip-SJ 97 7 4 3 5 8

(53%) 2 96 5 2 3 3 1

Cent

ral

Area

Co-CP 97 7 7 0 4 7

(58%) 2 96 8 7 1 3 1

FM-VB 97 9 7 2 3 6

(35%) 3 96 7 4 3 4 2

Sou

th

Wes

t

FL-TS 97 3 3 0 1 5

(50%) 1 Total 54 41 13 26 13

Source:Accident Records- Police of the State of São Paulo 4.1. Data Gathered at the Studied Crossings As previously mentioned, two sources of data were used in the study:

- data on pedestrian accidents, used for the determination of accident to (million) conflicts ratios and as the validation criterion on the detection of unsafe sites based on conflict counts;

- data on pedestrian-vehicle conflicts, used for the determination of the count limits for detecting sites with abnormal level of conflicts and as the diagnostic tool on the detection of unsafe sites based on conflict counts.

Details on the gathering and processing of data are shortly described in the following topics. a. Data on Pedestrian Accidents at the Studied Crossings Accident data were recovered from police records and police reports for two years before the conflict study (1996 and 1997). It was impossible to recover some of the accident reports of each intersection. Table 4 summarizes data for all the registered accidents and for recovered accident reports. Accidents referred to in Table 3 are only those with recovered reports and that occurred during the standard period of workdays (the same period of reference used as the standard period for expanded conflict counts). As can be seen, 26 accident reports were recovered from 54 registered pedestrian accidents and, among them, only 13 from 29 on the standard period of workdays. The accidents with recovered police reports were used in all analysis, except when estimating ratios of accidents to conflicts, for which the total number of accidents in the standard period of workdays was adjusted by the ratio of the known total to the recovered total, as described in the following, knowing that 29 accidents, instead of 13 only, occurred in the standard period of workdays (other accidents are not included). For each accident with a recovered accident report, an accident location index (%) had to be defined due to the lack of precise data on some police reports (see the column corresponding to accidents in Table 3). Most accidents were clearly located (i.e. had a 100% location index) but 6 of 13 pedestrian accidents had from 75/25 to 50/50 as accident location index (split between two



24

crossings, usually adjacent and/or similar) because pedestrian and vehicle movements were not clearly identified in most police accident reports (except when there was an obvious pattern commanded by the intersection layout and/or crossing site). Nevertheless, it was possible to verify that there was sites with pedestrian accidents among entry (near) crossings with through flow (as 1:Co-TA-BC, 1:FM-TP-CB e 3:Ip-E-TP) and among exit (far) crossings with turn flows (as 2:FL-TA-IP e 7:MA-TA), most of them with clear location. The accident location index was interpreted as the probability of occurrence of the accident in each crossing and the aggregate accident index was taken as the empirical estimate of the expected number of accidents in each crossing. The validation criterion for application of the TCT took the crossings with non-null accident index as unsafe and the others as safe (i.e. it is supposed that missing accidents occurred in the same crossings and that accident records were able to identify all the risky crossings). The suppositions described above would not be required when using better data (an unavoidable problem in our case study). b. Data on Pedestrian Conflicts at the Studied Crossings Conflict counts were carried-out during two days (a Monday and a Tuesday) of march, 1998 (actually, 16 and 17 of march), from 07:30 to 18:30 (using 6 one hour count and half hour rest periods, separated by one and a half hour for lunch just before or after mid-day, in each day, and amounting to 312 hours of observation on all crossings). Both days had good weather (and dry pavement). Recorded counts were expanded to the standard period of each day, following U.S.FHWA recommendations (PARKER/ZEGEER/1989a), and the average daily counts were obtained for each crossing and subsequently used in all analyses. Table 5 summarizes the data obtained for each crossing. One can note that sites with higher pedestrian-vehicle daily counts are not always the sites with pedestrian accidents (entry or near crossings with through flow, as 1:Co-TA-BC, 1:FM-TP-CB e 3:Ip-E-TP, and exit or far crossings with turn flows, as 2:FL-TA-IP e 7:MA-TA). This observation stresses the need of diagnostic parameters for selecting unsafe sites with better success. Table 5 also shows the main candidate variables for segmentation of crossing types (in column Seg.I and column Seg.II) as will discussed in the following. Conflict types were also similarly analyzed for aggregation.



25

Table 5 – Summary of average daily conflicts (for the standard 11 hours period), São Paulo Study. Basic Disaggregate Standard Period Conflict Count Data Agreggate

Inter. Crossing Seg.I Seg.II P/TPD P/TPE P/TAD P/TAE P/TDF P/TDR P/TEF P/TER P/TOT CO-CP 2: Co-TA-BC TA Ped - - 0.0 0.0 0.0 0.0 - - 0.0 CP-CM 6: CM-TA TA Ped - - 1.0 1.0 7.5 9.8 - - 19.3 CP-MA 7: MA-TA TA Ped - - 5.0 3.9 5.6 9.8 - - 24.3 FL-TS 4: FL-TA-PI TA Ped - - 3.7 3.7 - - - - 7.4 FL-TS 6: TS-E3-TA TA Ped - - 1.0 1.0 - - - - 2.0 FL-TS 7: TS-TA TA Ped - - 37.9 19.2 36.6 16.6 - - 110.3 CO-CP 4: Co-TA-CB TA Ped+ - - 0.0 1.0 - - 12.9 17.7 31.6 Ip-SJ 2: SJ-TA TA Ped+ - - 7.5 9.0 32.4 31.6 - - 80.5 Ip-SJ 5: Ip-E-TA TA Ped+ - - 9.2 9.6 - - 20.2 9.8 48.8 Ip-SJ 6: Ip-D-TA TA Ped+ - - 1.2 8.0 - - 12.9 10.5 32.6 FL-TS 2: FL-TA-IP TA Ped+ - - 2.7 6.1 - - 62.4 62.6 133.8 FM-VB 3: FM-TA-CB TA Ped+ - - 6.3 5.9 - - - - 12.2 FM-VB 6: VB-TA-CB TA Ped+ - - - - 7.9 16.5 - - 24.4 CO-CP 3:Co-TP-CB TP NSat 1.0 0.0 - - - - - - 1.0 FM-VB 4:VB-D-TP-BC TP NSat 0.8 1.7 - - - - - - 2.5 Ip-SJ 1: Sj-TP TP NSat 6.1 12.9 - - - - - - 19.0 Ip-SJ 3: Ip-E-TP TP NSat 9.9 2.7 - - - - - - 12.6 Ip-SJ 4: Ip-D-TP TP NSat 8.5 2.2 - - - - - - 10.7 FL-TS 1: FL-TP-IP TP NSat 1.0 6.2 - - - - - - 7.2 FL-TS 3: FL-TP-PI TP NSat 9.3 0.7 - - - - - - 10.0 FL-TS 5: TS-E2-TP TP NSat 5.5 11.2 - - - - - - 16.7 FM-VB 2: FM-B-TP-BC TP NSat 0.0 0.8 - - - - - - 0.8 CO-CP 1: Co-TP-BC TP Sat 0.0 0.7 - - - - - - 0.7 CO-CP 5: CP-TP TP Sat 0.0 0.0 - - - - - - 0.0 FM-VB 1: FM-TP-BC TP Sat 1.8 9.2 - - - - - - 11.0 FM-VB 5:VB-TP-BC TP Sat 0.8 2.4 - - - - - - 3.2

Note: Cells with – can not have conflicts of the related type. Zero counts are explicitly indicates as 0,0 (no count during the surveyed and standard periods). 4.2. Development of Parameters for the TCT Application to Pedestrian Safety Studies The development of parameters for the application of the TCT in the safety analysis with pedestrian-vehicle conflicts based on the U.S.FHWA guide includes:

- selecting a relevant segmentation of crossings; - selecting a final classification of conflict types; - determining limit counts for selected confidence levels; - determining accident to (million) conflicts ratios.

For the development of limit counts, the decision based criterion is straightly applied as there is a clear recommendation of selecting sites for treatments using on the comparison of average field counts to the established count limits, with a given confidence level, for all the conflict types discriminated on the type of crossing of each site. The decision based criterion is used to select the best segmentation of crossings, classification of conflict types and confidence level. The net economic index had to be computed with the sample of recovered accidents. With economic weights suggested in Table 2 (i.e., a 50% reduction potential with a unit cost of US$ 20000 per accident and a treatment cost of US$ 5000 per crossing), the actual “do nothing” or current situation can be associated to a loss of US$ 130000 related to the potential direct cost reduction (50%) for 13 accidents. The treatment costs amount to US$ 60000 in all the 12 accident sites (and US$ 70000 in all the 14 non-accident sites). Then, the lower and upper thresholds against which the net economic index of alternative TCT detections should be compared are:



26

- full success ME with treatment of the 12 sites with accidents only and a net benefit of plus US$70000 (savings of US$ 130000 less the cost US$ 60000 in 12 sites) and

- full failure WE with treatment of the 14 sites without accidents and a net benefit of minus US$ 70000 (avoidable accident costs are US$ 130000 as in the “do nothing”).

Note that relative values (RE) against the maximum (ME of US$ 70000) should be less sensible to benefit and cost weights for the comparison of alternative diagnostic criteria. Absolute values (DE) are more sensitive and should be taken as conservative ones in the subsequent discussion. For the development of accident to (million) of conflicts ratios, the conventional procedures for establishing parameters has to be used, as there is no recommended decision associated to the use of the risk measure. Then, a clear comparison of the alternative procedures can be made. a. Determination of Count Limits for Daily Conflicts Count limits C were determined for several segmentations, constrained by data availability. It is easy to see in Tables 3 and 4 that detailed segmentation of crossings is impossible with our data as the sample size in each segment quickly becomes too small and the validation analysis loses interest (as all the sites have no accident in the sample of some segments). The raw data on conflict counts is used as summarized in Table 5. The main segmentation criterion for the type of crossing distinguishes TP (i.e. entry, near or stop line crossings) and TA (exit, far or free crossings) sites. Remembering that all the carriageways are one-way, this is a fundamental classification based on pedestrian vehicle interaction and signs one version of the general distinction between protected and concurrent crossings as well. On TP crossings of signalized intersections, conflicts can occur only on traffic violations by pedestrians or vehicles and this possibility increases sharply when there are available gaps on vehicular flow. Then, further segmentation was considered based on the level of saturation of the vehicle approach (TP-Sat or TP-NSat). On TA crossings, there are concurrent movements of pedestrian and vehicles even without violations and conflicts can occur also in this situation. Then, segmentation based on pedestrian flow was analyzed because some crossings have very high traffic of pedestrians (TP-Ped or TP-Ped+, taking 900 ped/h as the threshold of high pedestrian flow crossings). In the following, the basic terminology used in the classification of conflict types or segmentation of crossing types is: aggregate/disaggregate for pooled or distinct groups of conflict types and complete/segmented for pooled or distinct groups of crossing types. Results are presented for the complete sample (with aggregate conflicts or disaggregate analyses by conflict type with 2, 4 and 8 types) and also for the TA/TP segmentation (with aggregate conflicts or disaggregate analyses by conflict type with 2 and 4 types). The category of each crossing for the main segmentation can be seen checking the column Seg.I of Table 5. Both analyses are carried-out for three levels of confidence (75%, 90% and 95%). More refined segmentations, with TP-Sat/TP-NSat and TP-Ped/TP-Ped+ were also submitted to a preliminary study of aggregate and disaggregate analyses. The refined category of each crossing can



27

be checked on the column Seg.II of Table 5. Despite the small samples available in each cell, the preliminary study of more detailed data can suggest the potential gain from further segmentation (and its effect on the need of a disaggregate analysis of conflict types). Table 6 summarizes count limits based on the sample of crossings, taking several alternatives for segmentation of crossings and disaggregation of conflict types. Part 6a contains the basic data for the complete sample and the segmentation of crossings in two groups TA/TP, with 4 types (P/TP, P/TA, P/TD and P/TE), with 2 types (P/VA, with P/TP and T/TA, and P/VT, with P/TD and PTE) or pooling all conflicts in an aggregate type. Part 6b contains preliminary data on more detailed analysis with further segmentation (TA-Ped/TA-Ped+, TP-NSat/TP-Sat) or disaggregation (all the 8 conflict types of Figure 1). Based on the count limits, the diagnosis of safe or unsafe site is carried-out for the crossings and compared to the accident records. Note that the use of the accident location index spreads accidents among crossings but acts in the opposite direction of using the diagnosis based on the sample of recovered police accident reports only. Alternative analyses with rounding of the cumulative accident index of each site delivered comparable results and the analysis based on a 50/50 split of accidents (i.e. with less precise location) were kept for presentation. The main result of the paper, in Table 7, presents the evaluation index of each alternative segmentation and disaggregation and also for each of the 3 confidence levels. The indexes were evaluated using the representative weights of Table 2. The nomenclature for errors type I and II or hits type I and II are also the same of Table 2 as well. The lower confidence level was introduced based on the observation that when higher accident savings are conjectured the gain achieved is very significant (i.e. a more stringent statistical criterion runs against safety). Based on our data, the use of the 75% confidence level with the TA/TP segmentation and aggregate conflicts is clearly preferred (the TA/TP segmentation and the classification with 4 or 8 conflict types is also recommended). The compromise between Type I and Type II hits (or errors) is clearly shown in Table 7 when comparing the %Hit1 and %Hit2 (the preferred alternative reaches 41.67% hits on identifying accident sites and 71.43% hits on identifying safe sites). The effect of requiring higher confidence levels is clearly displayed (increase of %Hit2, identifying safe sites, at the expense of decreasing %Hit1, identifying unsafe sites). Note that the TP/TA segmentation is firmly supported for all disaggregation of conflict types, being essential for achieving good results when using up 2 conflict types. The TP/TA segmentation is implicit in the use of 4 or 8 types but the interesting point here is that the use of more refined conflict types can change the pattern of errors by favoring safety. For example, with the 75% confidence level and the TA/TP segmentation, the use of 8 conflict types increases %Hit1 to 58.33% (detection of unsafe sites) and decreases %Hit2 to 64.29% (detection of safe sites).



28

Table 6 – Results on Normal Count Limits for Pedestrian-Vehicle Conflicts, São Paulo Study. a. Basic Results for Count Limits based on the Sample (Confidence Levels CL=75%, 90%, 95%). Segmentation All-

Aggregate All- 2Types

TA/TP-Aggregate

TA/TP- 2 Types

TA/TP- 4Types

Conflict Type Aggregate P/VA P/VT Aggregate P/VA P/VT P/TP P/TA P/TD P/TE

CrossingType All All TA TA All-TA

average 23.9 9.6 38.3 40.6 12.0 38.3 12.0 29.1 52.3 variance 1162.1 135.6 1263.9 1781.5 237.7 1263.9 237.7 598.9 2362.9 CL=75% 31.6 13.1 53.0 56.2 16.3 53.0 16.3 39.9 72.3

90% 65.0 24.2 85.0 95.2 31.1 85.0 31.1 61.5 116.1 95% 92.4 33.0 108.9 125.0 43.0 108.9 43.0 77.3 148.8

CrossingType TP TP All-TP

average 7.3 7.3 7.3 variance 42.0 42.0 42.0 CL=75% 10.1 10.1 10.1

90% 15.9 15.9 15.9 95% 20.2 20.2 20.2

b. Further Results for Count Limits based on the Sample (Confidence Levels CL=75%, 90%, 95%). Segmentation Ped/+/N/S

Aggregate Ped/+/N/Sat- 2 Types

TA/TP- 8Types

Conflict Type Aggregate P/VA P/VT P/TPD P/TPE P/TAD P/TAE P/TDF P/TDR P/TEF P/TER

CrossingType TA-Ped TA-Ped All-TA

average 27.2 12.9 21.5 6.3 5.7 15.0 14.1 27.1 25.2 variance 1748.7 480.8 507.3 108.3 29.1 237.9 110.8 565.7 636.1 CL=75% 34.8 15.6 29.8 7.7 7.9 20.8 19.1 37.3 34.9

90% 76.1 37.3 50.6 18.0 12.8 35.0 28.1 58.5 58.0 95% 110.9 56.3 66.5 27.0 16.4 45.8 34.6 74.1 75.5

CrossingType TA-Ped+ TA-Ped+

average 52.0 11.1 49.6 variance 1775.3 40.1 1591.8 CL=75% 71.1 14.5 67.8

90% 107.9 19.6 102.6 95% 134.8 23.1 127.9

CrossingType TP-NSat TP-NSat All-TP average 8.9 9.9 3.4 3.9 variance 44.1 40.3 14.8 20.0 CL=75% 12.1 13.2 4.8 5.4

90% 17.8 18.4 8.4 9.6 95% 21.9 22.1 11.1 12.9

CrossingType TP-Sat TP-Sat average 3.7 3.7 variance 25.4 25.4 CL=75% 5.0 5.0

90% 9.9 9.9 95% 13.9 13.9



29

Table 6 – Comparative Performance of Alternative Criteria for Detection of Abnormal Conflict Level.

Confidence Level

All- Aggregate

TA/TP- Aggregate

Ped/+/N/Sat- Aggregate

All- 2Types

TA/TP- 2 Types

Ped/+/N/Sat- 2 Types

TA/TP- 4Types


TA/TP- 8Types


75% NoErr1 10 7 8 10 7 8 7 9 5 6 NoHit1 2 5 4 2 5 4 5 3 7 6

%Hit1 16.67% 41.67% 33.33% 16.67% 41.67% 33.33% 41.67% 25.00% 58.33% 50.00%

NoErr2 3 3 3 4 4 4 4 4 5 5 NoHit2 11 11 11 10 10 10 10 10 9 9

%Hit2 78.57% 78.57% 78.57% 71.43% 71.43% 71.43% 71.43% 71.43% 64.29% 64.29% AccE1 11 7 7,5 11 7 7,5 7 9 5,5 6,5 AccH1 2 6 5,5 2 6 5,5 6 4 7,5 6,5

DE -5 20 20 -10 15 15 15 5 15 10 TA-Ped -5 -5 -5 -5 -5 -5 -5 -5 -5 -5 TA-Ped+ 0 10 10 5 5 5 5 5 0 0 TP-NSat 0 0 0 -10 0 0 0 -10 5 0 TP-Sat 0 15 15 0 15 15 15 15 15 15

RE -7.14% 28.57% 28.57% -14.29% 21.43% 21.43% 21.43% 7.14% 21.43% 14.29%

90% NoErr1 10 11 10 11 11 10 10 10 7 10 NoHit1 2 1 2 1 1 2 2 2 5 2

%Hit1 16.67% 8.33% 16.67% 8.33% 8.33% 16.67% 16.67% 16.67% 41.67% 16.67%

NoErr2 1 3 2 1 3 2 3 2 3 3 NoHit2 13 11 12 13 11 12 11 12 11 11

%Hit2 92.86% 78.57% 85.71% 92.86% 78.57% 85.71% 78.57% 85.71% 78.57% 78.57%

AccE1 11 12 10 12 12 10 11 10 8,5 10 AccH1 2 1 3 1 1 3 2 3 4,5 3

DE 5 -10 10 0 -10 10 -5 10 5 5 TA-Ped -5 -5 -5 -5 -5 -5 -5 -5 -5 -5 TA-Ped+ 10 5 5 5 5 5 10 5 10 5 TP-NSat 0 -10 -5 0 -10 -5 -10 -5 0 -10 TP-Sat 0 0 15 0 0 15 0 15 0 15

RE 7.14% -14.29% 14.29% 0.00% -14.29% 14.29% -7.14% 14.29% 7.14% 7.14%95% NoErr1 11 11 12 11 11 12 12 12 12 12

NoHit1 1 1 0 1 1 0 0 0 0 0

%Hit1 8.33% 8.33% 0.00% 8.33% 8.33% 0.00% 0.00% 0.00% 0.00% 0.00% NoErr2 1 0 0 1 1 1 1 1 2 1 NoHit2 13 14 14 13 13 13 13 13 12 13

%Hit2 92.86% 100.00% 100.00% 92.86% 92.86% 92.86% 92.86% 92.86% 85.71% 92.86% AccE1 12 12 13 12 12 13 13 13 13 13 AccH1 1 1 0 1 1 0 0 0 0 0

DE 0 5 0 0 0 -5 -5 -5 -10 -5 TA-Ped -5 0 0 -5 -5 -5 -5 -5 -5 -5 TA-Ped+ 5 5 0 5 5 0 0 0 0 0 TP-NSat 0 0 0 0 0 0 0 0 -5 0 TP-Sat 0 0 0 0 0 0 0 0 0 0

RE 0.00% 7.14% 0.00% 0.00% 0.00% -7.14% -7.14% -7.14% -14.29% -7.14%



30

The absolute or relative economic indexes are handsome in delivering an unambiguous selection. Note that DE>0 is achieved in all cases when using the 75% confidence level with the TA/TP segmentation or at least 4 conflict types. The use of the TA/TP segmentation (again implicit in the use of 4 or 8 conflict types) is clearly superior. Nevertheless, the relative index shows clearly that more insightful classification/segmentation is needed. For getting clues on improvements, it is interesting to use the fact that the DE can be (additively) decomposed in each sub-sample of crossing types. One can see that the net economic performance is worse in crossings of the TA-Ped and TP-NSat groups (suggesting where better data and segmentation would perhaps bring some additional precision in these classes). Some results not shown in Table 7 are also worth noting. During the evaluation, it became clear that the result on the option with the best performance index was sensible to the values of weight for most confidence levels. With representative values at the border of economic efficiency for treatments (an accident saving of US$ 7500 with the treatment cost of US$ 5000), the same pattern of results was obtained. Higher accident saving values will favor even more liberal criteria, on behalf of safety benefits. Then, the use of more stringent statistical criteria (as the 95% confidence level included in the U.S.FHWA guide, PARKER/ZEGEER/1989a) seems to be overly conservative on economic grounds in any reasonable context. Note that the use of the more detailed segmentation is promising, based on our results. The same pattern seems to be suggested by results from the more detailed disaggregation of conflict types but, in this case, better performance is limited to higher accident savings values, especially for higher confidence levels, what seems to be due to the selection of a larger number of sites for treatment (nevertheless, the same can be more conveniently achieved by lowering the required confidence level). b. Determination of Ratios of Accidents per Million Conflicts. The difficulty in identifying movements of vehicles and pedestrians involved in traffic accidents commanded an aggregate analysis of the risk measure based on the ratio of accidents per million conflicts. The following analysis is, then, limited to the comparison of the ratio on each crossing type (all, TA/TP, TA-Ped/TA-Ped+ and TP-NSat/TP-Sat). The conversion of daily conflicts in the standard period to annual conflict totals is carried-out using representative factors as displayed in Table 8 (suggested in PARKER/ZEGEER/1989a and recommended in PIETRANTONIO/1991 also). As our results are limited to workdays and dry pavements, the ratios can also be calculated as a rough ratio RR of yearly accidents per hourly conflict. Despite its simplicity, we applied this procedure previously with good practical results. Table 8 – Representative Factors for Converting Standard Period Conflict Counts to Annual Counts.

dry pavement wet pavement Std/Day Days/Year Std/Day Days/Year

Work Days 70% 4/7 of 365 70% 1/7 of 365 Week Ends 70% 3/14 of 365 70% 1/14 of 365



31

With such conversion factors, the calculation of the annual frequency of traffic conflicts is straightforward and can be related to the annual frequency of accidents (given our two-year data samples). Nevertheless, as our sample of accidents was only able to recover data on 13 police reports of 29 occurrences, we had to eliminate this partial data bias by inflating the estimated ratio by a correcting factor (29/13=2.23). One should note that this correction does not account for partial record of occurrences (that would also be present in our data as usual). Table 9 summarizes the results on the ratio of accidents per million conflicts for each of the pedestrian crossing segments used in this study. Part 9a shows that the difference of the ratio of the risk measure of conflicts in TA and TP crossings is suggested to be highly relevant (1 to 8 or one order of magnitude) and also statistically significant (at least based on the approximate test on the quasi-t statistic). A more detailed analysis is shown on Part 9b. At a smaller degree, the same confidence can be attributed to the difference between the risk measures of conflicts on TP-NSat and TP-Sat crossings and clearly point to the importance of diagnostic parameters for a proper analysis. Despite the high variance, only the segmentation of TA crossings based on pedestrian flows is discarded as non-relevant and non-significant. This conclusion can be attributed to the high level of pedestrian flow (900 ped/h) used as a threshold between classes. As our sample have a very small number of crossings with small pedestrian flows, our choice was constrained. One can clearly see that differences even when the values of the ratio are relevant based on an engineering criteria but the statistical significance is reduced by the high variance of the estimates. Use of empirical conversion factors in the expansion of daily conflict into yearly million of conflicts could improve the forecasting of accidents but also could increase the variance of estimates (HAUER/1997, chapter 8). For example, the difference of the ratio on TP-NSat and TP-Sat crossing types is relevant yet (almost 1 to 4), despite less significant statistically. The comparison of these two segments of crossing types is also illustrative of the meaning of the diagnostic parameters. Note that the risk measure of conflicts is higher in crossings of the TP-Sat type (in stop lines of saturated vehicular approaches). Nevertheless, the frequency of conflicts is higher in crossings of the TP-NSat type (almost 2 to 1, as shown by the average count in Table 6b). Then, crossings of the TP/NSat type were predicted as finally safer (taking the product of each measure).



32

Table 9 – Results on the Ratio of Pedestrians Accidents to Million Conflicts, São Paulo Study. a. Basic Results for the Ratio of Accidents to Million Conflicts based on the Sample.

Aggregate- All

Aggregate- TA

Aggregate- TP

diff: TA/TP

No.Sites 26 13 13 Total Accidents 13.0 5.5 7.5 R (Ac/MC) 35.0387 17.5065 131.9247 -114.4182 Standard Dev. 12.7482 8.0006 51.0302 51.6536 Quasi-t 2.7485 2.1882 2.5852 -2.2151 Coef.Variation 36.38% 45.70% 38.68% Adjusted Ratio * 78.1632 39.0531 294.2936 RR (Ay/Ch) 0.2562 0.1280 0.9646 MeanAccidents 0.50 0.42 0.58 MeanAbsErrorR ** 0.67 0.47 0.61 MaxAbsErrorR ** 2.30 1.75 1.49 MeanAbsErrorS ** 0.54 0.52 0.56 MaxAbsErrorS ** 1.50 1.58 1.42

* The adjusted ratio is the final parameter for practical purposes as it includes a general correction factor (actually 29/13=2.23), for the partial set of accident reports recovered and used in the analysis. ** ErrorR and ErrorS (Mean or Max) are errors of predicted against observed accident frequencies using the ratio of accidents to conflict counts and the sample average of accident counts, respectively. (R is in Ac/MC=accidents per million of conflicts; RR is in Ay/Ch=yearly accidents per hourly conflicts) b. Further Results for the Ratio of Accidents to Million Conflicts based on the Sample.

Aggregate- TA-Ped

Aggregate- TA-Ped+

diff:TA- Ped/Ped+

Aggregate- TP-NSat

Aggregate- TP-Sat

diff:TP- NSat/Sat

No.Sites 6 7 9 4 Total Accidents 2.5 3.0 4.5 3.0 R (Ac/MC) 25.6902 13.8341 11.8560 93.8058 337.8689 -244.0631 Standard Dev. 26.9684 6.9915 27.8599 43.8764 142.4120 149.0179 Quasi-t 0.9526 1.9787 0.4256 2.1380 2.3725 -1.6378 Coef.Variation 104.98% 50.54% 46.77% 42.15% Adjusted Ratio * 57.3089 30.8608 209.2592 753.7075 RR (Ay/Ch) 0.1878 0.1011 0.6859 2.4703 MeanAccidents 0.42 0.43 0.50 0.75 MeanAbsErrorR ** 0.67 0.32 0.52 0.43 MaxAbsErrorR ** 1.69 0.80 1.06 0.86 MeanAbsErrorS ** 0.56 0.49 0.44 0.75 MaxAbsErrorS ** 1.58 0.57 1.00 1.25

* The adjusted ratio is the final parameter for practical purposes as it includes a general correction factor (actually 29/13=2.23), for the partial set of accident reports recovered and used in the analysis. ** ErrorR and ErrorS (Mean or Max) are errors of predicted against observed accident frequencies using the ratio of accidents to conflict counts and the sample mean of accident counts, respectively. (R is in Ac/MC=accidents per million of conflicts; RR is in Ay/Ch=yearly accidents per hourly conflicts)



33

A last point can be made about the precision of estimates of the expected accident level based on the use of the accident to (million) of conflict ratio. As previously mentioned, the comparison with predicted and observed counts can reveal the explained and unexplained error achieved with different methods and should be weight against the specific content of each method (as the safety diagnostic based on the direct observation of traffic operations that the TCT is able to carried-out but can not be achieved by most alternative methods). Just for accident prediction, the different methods can be assessed as competitors and, as an example, be embodied into an Empirical Bayesian procedure (the currently preferred statistical evaluation method in safety research). In this case, the prior expected accident frequency can be estimated from accident counts on sample of similar sites, from exposure or conflict based ratios or from statistically calibrated model equations (a more sophisticated technique that can use several explanatory variables, that in the other techniques are usually considered only in developing segmentation alternatives). Despite originated by the need to taking into account the regression to the mean effect (see HAUER/1997), the EB approach can be used as a variance reduction tool with all the other methods. Table 9 includes data from the absolute average and maximum errors achieved with accident prediction on the samples based on the use of the ratio estimators and the accident counts on each the crossing segmentation considered. With accident counts, the average accident frequency in each segment was used as predictor (corresponding to the prediction based on the sample mean). Based on their estimating equations (see CAMERON/TRIVEDI/1998, item 3.2), it can be shown that the same predictors are generated by a poissonian or negative binomial regression model for accident frequency with group specific dummies as explanatory variables. As can be seen, average and maximum absolute errors are large and similar for both methods but more dependent on the segmentation for the ratio estimates (at least without conflict types). Then, more sophisticated models/refined segmentations is a need but the precision achieved with the ratios is comparable to similar simple models based on accidents.



34

5. CONCLUSIONS This research aimed at determining the diagnostic parameters for the analysis of pedestrian safety problems at signalized intersections based on the traffic conflict analysis technique (TCT), using the U.S.FHWA concepts for refined diagnosis. Before undertaking our work, we discussed specific criteria for observing and analyzing pedestrian-vehicle conflicts, trying to avoid the hidden supposition of a simple and direct analogy to vehicle conflicts, and discussing specific problems related to the typology of conflicts and crossings. The methods used for the determination of diagnostic parameters were reviewed and a new decision based criterion for tracking the practical performance of their use in applied work was proposed as a complement to conventional statistical criteria. Both approaches were applied to a case study using a sample of crossings with high vehicular and pedestrian flows of the City of São Paulo, Brazil, for illustrative purposes. The main result obtained was the support to the segmentation of entry (near or stop-line) crossings (TP) and exit (far or free) crossings (TA). Given our sample of divided carriageway or one-way approaches, this is clearly sound and would suggest that other classes could be relevant in a larger sample (crossings on two-way approaches and unsignalized crossings at least). For establishing limit counts, the parameters used as comparator in selecting sites for treatment based on their field data, the proposed method was shown to be able to select the best segmentation and confidence levels for count limits to be used in practical work. The criterion has a clear economic interpretation and can be decomposed additively so as to reveal sub-samples where achievements are better or worse. The results on the criterion for identifying abnormal conflict counts recommended the use of a 75% confidence level and at least the segmentation of near or stop-line crossings (TP) and far or free crossings (TA). The use of 4 conflict types (the same types used in the U.S.FHWA guide) or 8 conflict types are also suggested on a preliminary analysis but more data is need to confirm that further benefits can be expected from more detailed parameters. Nevertheless, the study showed that the decision on the best criteria is sensitive to benefit and cost parameters, what can constraint the scope for detailing. For accident to (million of) conflicts ratio, conventional statistics were used for the same alternative groups of crossings as there is no clear decision implied by the ratios or by the accident predictions based on their use. For the measure of risk (accident proneness), constrained by the impossibility of a precise identification of vehicle and pedestrian movements involved in accidents, a relevant and significant difference between the ratio of accidents per million conflicts of near or stop-line crossings (TP) and far or free crossings (TA) were the only result. This lack of resolution can be partially attributed to the intrinsic variance of accident and conflict data. As shown, despite displaying relevant differences on the risk level of conflicts, the results were buried by their high variance and just suggested the potential of more detailed segmentations based on alternative criteria (saturation level on TP crossings and, less clearly, of pedestrian flows on TA crossings, as the sample was restricted to crossings with large pedestrian traffic).



35

The main result obtained, however, is the demonstration that the decision based criterion was very useful for selecting the segmentation of crossings and/or classification of conflict types and for justifying the best confidence level to be used in professional studies. Further research can extend the method to the development of parameters for other techniques and can also improve the method by including other effects in addition to accident reduction. Despite the need of further improvements, at least the basic results seem to be preliminarily applicable to the diagnosis of vehicle-pedestrian conflicts on the type of crossings studied. It is important to note that traffic conflicts at TP crossings, especially at TP-NSat crossings, are very rare events and could easily be outside the scope of safety problems that can be firmly evaluated with TCT studies. Despite being more usable than accident data for the evaluation of countermeasures in a safety studies to be conducted shortly after their implementation, warranting the maintenance of the relevant site features, the limitations of TCT data have also to be clearly understood (as in the evaluation of rare conflicts). Complementary methods and data should be searched for analyzing less frequent events that appear between vehicles and pedestrians (as between vehicles). Road Safety Audits could be more practical in the diagnostic of safety problems with this kind of events. Conflict Opportunity Measures could a replacement for traffic conflicts in these settings as well as can be useful as a prediction rod for count limits or other safety parameters. This option can be an interesting alternative to the need of extensive data gathering and processing efforts for each new setting and for developing parameters that can embody the effect of road traffic and control variables, among others. This is our current undertaking.



36

Acknowledgements: especially to FAPESP-Foundation for Research Aid of the State of São Paulo and to CET/SP-Traffic Engineering Company of the City of São Paulo, for their collaboration in the original 1998 study, that gathered the data use in the research. Of course, all statements and remaining errors are our only and own responsibility. REFERENCES ABDULSATTAR, N.H.; TARAWNEH, S.M.; MCCOY, P.T. AND KACHMAN, S.D. (1996) - Effect on Vehicle-

Pedestrian Conflicts of ´Turning Traffic Must Yield to Pedestrians´ Signs. Transportation Research Record, n.1553, TRB, USA.

ALMQVIST, S. (1998) - Introduction of the Swedish Traffic Conflict Technique – a Method for Assessing Traffic

Safety. FAPESP-Fundação de Amparo à Pesquisa do Estado de São Paulo. (Relatório e Anexos Ia IV, Processo 97/12.154-0), São Paulo, Brazil.

ALMQVIST, S. E HYDÉN, C. (1994) - Methods for Assessing Traffic Safety in Developing Countries – Lund:

Lund University. (Building Issues – Lund Centre for Habitat Studies, Volume 6) BAGULEY, C. (1988) - Guidelines for the Traffic Conflict Technique. Hertford: IHT– Institute of Highway

Transportation. (University College London Transport Studies Group) CAMERON, A. C.; TRIVEDI, P.K. (1998) –Regression Analysis of Count Data, Cambridge University Press. CLARK, K.L.; HUMMER, J.E.; AND DUTT, N. (1996) - Field Evaluation of Fluorescent Strong Yellow-Green

Pedestrian Warning Signs. Transportation Research Record, n.1538, TRB, USA. COCHRAN, W. G. (1977) - Sampling Techniques. New York: John Wiley & Sons. Third Edition. CYNECKI, M.J. (1980) - ‘Development of a Conflicts’ Analysis Technique for Pedestrian Crossings. Transportation

Research Record, n.743, TRB, USA. DAVIS, S.E.; ROBERTSON, H.D. E KING, L.E. (1989) - Pedestrian/Vehicles Conflicts: An Accident Prediction

Model. Transportation Research Record, n.1210, TRB, USA. GARDER, P. (1989) - Pedestrian Safety at Traffic Signals: A study Carried Out with the Help of a Traffic Conflicts

Technique. Accident Analysis & Prevention, Great Britain, vol. 21, n.5, p.435-444. GLAUZ, W.D.; BAUER, K.M.; & MIGLETZ, D.J. (1985) - Expected Traffic Conflict Rates and Their Use in

Predicting Accidents. TRB, Transportation Research Record, USA, n.1026. GLAUZ, W.D. & MIGLETZ, D.J. (1980) - Application of Traffic Conflict Analysis at Intersections. TRB, NCHRP

Report, USA, n.219. GUEDES, E.; BRAGA, M.; PIETRANTONIO, H. (1997) – Initial Experiences with Traffic Conflict Technique in

Brazil. Proceedings of the ICTCT97 Conference, Lund, Sweden. HAUER, E. (1975) - The Traffic Conflict Technique – Fundamental Issues. Publication 75-01. Department of Civil

Engineering, University of Toronto, Ontario, Canada. HAUER; E. (1997) –Observational Before/After Studies in Road Safety: Estimating the Effect of Highway and

Traffic Engineering Measures on Road Safety, Pergamon Press. HUMMER; J.E. (1994) - Traffic Conflict Studies, chapter 12 in Robertson, H.D.; Hummer, J.E.; Nelson, D.C (eds.),

Manual of Traffic Engineering Studies, ITE, USA.



37

HYDÉN, C. (1987) - The Development of a Method for Traffic safety Evolution: The Swedish Traffic Conflicts

Technique. Bulletin 70, Sweden. ICTCT - INTERNATIONAL COMMITTEE ON TRAFFIC CONFLICT TECHNIQUE (1984) - The Malmö Study.

Netherlands: Institute for Road Safety Research SWOV. JAVID, M. AND SENEVIRATNE, P.N. (1991) - Applying Conflict Technique to Pedestrian Safety Evaluation. ITE

Journal, Washington D.C., p.21-26. LAW, A.M.; KELTON, W.D. (1991) – Simulation Modeling & Analysis, McGraw-Hill USA. LORD, D. (1994) - Analysis of Pedestrian Conflicts with Left-Turning Traffic. Transportation Research Record,

n.1538, TRB, USA. MUHLARD, N. (1988) Technique des Conflicts de Trafic – Manuel des L’Utilisateur, Synthese INRETS –

Institute National de Recherche sur les Transports et leur Securite n.11, France. PARKER Jr., M.R.; ZEGEER C.V. (1989a) - Traffic Conflict Techniques for Safety and Operations – Engineering

Guide. Publication FHWA-IP88-026, FHWA - Federal Highway Administration, Department of Transportation, U.S.A.

PARKER Jr., M.R.; ZEGEER C.V. (1989b) - Traffic Conflict Techniques for Safety and Operations – Observer

Manual. Publication FHWA-IP88-027, FHWA - Federal Highway Administration, Department of Transportation, U.S.A.

PIETRANTONIO, H. (1991) Manual de Procedimentos para Análise de Conflitos de Tráfego em Interseções. IPT

- Instituto de Pesquisas Tecnológicas do Estado de São Paulo. Agrupamento de Engenharia de Transportes, Divisão de Tecnologia de Transportes, São Paulo, Brazil (in portuguese, avalable at www.poli.usp.br/p/hugo.pietrantonio).

PIETRANTONIO H. (1999) - Avaliação da Técnica Sueca de Análise de Conflitos de Tráfego – Aplicação ao

Estudo de Problemas de Segurança de Pedestres em Interseções Semaforizadas da Cidade de São Paulo. EPUSP - Escola Politécnica da Universidade de São Paulo. Working Report LEMT No.2/98, São Paulo, Brazil (in portuguese, avalable at www.poli.usp.br/p/hugo.pietrantonio).

ROBERTSON, H.D. (1977a) Pedestrian Signal Displays: An Evaluation of Word Message and Operation.

Transportation Research Record, n.629, TRB, USA. ROBERTSON, H.D. (1977b) - Pedestrian Preference for Symbolic Signal Displays. Transportation Engineering, ITE,

USA. SAYED, T.; ZEIN, S. (1999) – Traffic Conflict Standards for Intersections. Transportation Planning and

Technology, vol.22, n.4, pp.309-323. U.S.Department of Transportation. (1981) - Highway Safety Engineering Studies Procedural Guide. USA: Federal

Highway Administration. ZEGEER, C.V. (1991) Synthesis of Safety Research – Pedestrians. Publication FHWA-SA-91-034, USA, FHWA -

Federal Highway Administration, U.S. Department of Transportation, USA. ZEGEER, C.V.; RANDOLPH D.A.; FLAK M.A. & BHATACHARYA R.K. (1980) Use of Pedestrian Conflict

Analysis for the Hazard Assessment in School Zones. Transportation Research Record, n.743, TRB, U.S.A. ZEGEER, C.V.; OPIELA, K.S.; CYNECKI, M.J. (1982) - Effect of Pedestrian Signals and Signal Timing on Pedestrian

Accidents. Transportation Research Record, n.848, TRB, U.S.A.



38

APPENDICES



39

A. Forms and Criteria of the U.S.Guide on the Traffic Conflict Technique (of the FHWA-

Federal Highway Administration and the ITE-Institute of Traffic Engineers)



40

Traffic Conflicts

Site: _______________________________________ Observer: _______________________________ Date: ________________ Day: __________________ Unit Period for Counting: _____________ min.

P=primary conflicts; s=secondary conflicts

Approach No.: ____

Comments

Time to Start the Counting

Period

Per.No

P s P s P s P s P s P s P s P s P s P s P s P s P s P s P s

Total P+s

General

Comments: _______________________________________________________________________________________

Figure A1. Form for Field Observation of Traffic Conflicts in the U.S.FHWA Guide (PIETRANTONIO/91, based on PARKER/ZEEGER/1989a)



41

Table A1. Required Hours for Observation of Traffic Conflicts on 4-leg Intersections, by Type of Control and Traffic Level

+=================================================================================================================+ | Signalized Intersections with 4 Approaches | |-----------------------------------------------------------------------------------------------------------------| | | 10000 to 25000 | More than 25000 | | | vehicles per day | vehicles per day | | Type of Conflict | Average Hourly Required Hours| Average Hourly Required Hours| | | Conflicts of Observation| Conflicts of Observation| |-------------------------------------------------|-------------------------------|-------------------------------| | Same Direction – Left Turn | 12.25 1.6 | 7.60 4.1 | | Same Direction – Slow Vehicle | 34.36 0.1 | 60.82 0.1 | | Same Direction – Lane Change | 0.69 * | 1.66 * | | Same Direction – Right Turn | 11.32 1.0 | 19.88 0.3 | | Same Direction – All types | 58.61 0.3 | 89.96 0.1 | | Opposed Left Turn | 2.64 1.8 | 2.00 3.2 | | Angle – Left Turn from Left | 0.04 * | 0.06 * | | Angle – Through from Left (near approach) | 0.03 * | 0.01 * | | Angle – Right Turn from Left | 0.03 * | 0.01 * | | Angle – Left Turn from Right | 0.05 * | 0.04 * | | Angle – Through from Right (far approach) | 0.02 * | 0.03 * | | Angle – Right Turn from Right | 0.34 * | 0.24 * | | Opposed Right Turn on Red | 0.01 * | 0.02 * | |=================================================================================================================| | Unsignalized Intersections with 4 Approaches | |-----------------------------------------------------------------------------------------------------------------| | | 2500 to 10000 | 10000 to 25000 | | | vehicles per day | vehicles per day | | Type of Conflict | Average Hourly Required Hours| Average Hourly Required Hours| | | Conflicts of Observation| Conflicts of Observation| |-------------------------------------------------|-------------------------------|-------------------------------| | Same Direction – Left Turn | 6.42 5.7 | 12.07 1.6 | | Same Direction – Slow Vehicle | 9.26 0.7 | 13.80 0.3 | | Same Direction – Lane Change | 0.01 * | 0.25 * | | Same Direction – Right Turn | 5.26 4.4 | 5.61 3.9 | | Same Direction – All types | 20.96 1.9 | 29.01 1.0 | | Opposed Left Turn | 0.33 * | 0.82 19.1 | | Angle – Left Turn from Left | 0.31 * | 0.36 * | | Angle – Through from Left (near approach) | 0.61 12.3 | 0.30 * | | Angle – Right Turn from Left | 0.05 42.0 | 0.02 * | | Angle – Left Turn from Right | 0.45 * | 0.39 * | | Angle – Through from Right (far approach) | 0.48 16.5 | 0.30 3.8 | | Angle – Right Turn from Right | 0.50 48.4 | 0.82 18.0 | +=================================================================================================================+ Notes: . * means that 2 or more weeks of observation are required, for normal levels of conflict frequency; . Required hours of observation to estimate the average hourly count with +/-50% of precision and 90% of significance; . Sampling of conflicts should be evenly split among the 4 approaches of a signalized intersection or among the 2 priority approaches of an unsignalized intersection. All counts for daytime and dry pavement conditions, without secondary conflicts.

(PIETRANTONIO/91, based on PARKER/ZEEGER/1989a)



42

Table A2. Standard Daily Conflict Counts for Average and Abnormal Level on Two-way Stop Controlled Unsignalized Intersections with 4 Approaches

+===================================================================================================================+ | Traffic Volume:2500 to 10000 Vehicles per Day (Total) Abnormal Daily Level | | of Conflict Count | | Type of Conflict Average Daily Variance 90th 95th | | Conflict Count Percentile Percentile | |-------------------------------------------------------------------------------------------------------------------| | 1.Same Direction – Left Turn | 70.645 1005.0 110.0 130.0 | | 2.Same Direction – Slow Vehicle | 101.861 9648.2 225.0 295.0 | | 3.Same Direction – Lane Chance | 0.105 0.050 - - | | 4.Same Direction – Right Turn | 57.912 2197.3 120.0 150.0 | | 5.opposed Left Turn | 3.640 8.300 7.5 9.0 | | 6.Angle – Left Turn from Left | 3.366 7.790 7.0 9.0 | | 7.Angle – Through from Left (near approach) | 6.698 42.0 1.5 19.0 | | 8.Angle – Right Turn from Left | 0.567 0.828 - - | | 9.Angle – Left Turn from Right | 4.493 72.7 16.0 23.0 | |10.Angle – Through from Right (far approach) | 5.228 11.6 10.0 12.0 | |11.Angle – Right Turn from Right | 5.546 12.1 10.0 12.0 | | All Same Direction (1 to 4) | 230.523 17929.2 410.0 490.0 | | All Angle – Through (7 + 10) | 11.926 75.2 24.0 29.0 | |===================================================================================================================| | Traffic Volume:10000 to 25000 Vehicles per Day (Total) Abnormal Daily Level | | of Conflict Count | | Type of Conflict Average Daily Variance 90th 95th | | Conflict Count Percentile Percentile | |-------------------------------------------------------------------------------------------------------------------| | 1.Same Direction – Left Turn | 132.745 11643.4 275.0 350.0 | | 2.Same Direction – Slow Vehicle | 151.831 5921.8 255.0 290.0 | | 3.Same Direction – Lane Chance | 2.797 22.6 - - | | 4.Same Direction – Right Turn | 61.695 1156.5 105.0 125.0 | | 5.Opposed Left Turn | 8.982 39.8 17.0 21.0 | | 6.Angle – Left Turn from Left | 3.913 6.452 7.0 9.0 | | 7.Angle – Through from Left (near approach) | 3.250 4.644 6.0 7.5 | | 8.Angle – Right Turn from Left | 0.165 0.077 - - | | 9.Angle – Left Turn from Right | 4.333 21.2 10.0 14.0 | |10.Angle – Through from Right (far approach) | 3.327 4.297 6.0 7.5 | |11.Angle – Right Turn from Right | 8.972 99.4 21.0 29.0 | | All Same Direction (1 to 4) | 319.068 28650.5 540.0 640.0 | | All Angle – Through (7 + 10) | 6.577 15.7 12.0 14.0 | +===================================================================================================================+ Notes: . Conflict Counts are the total number of conflicts on the standard 11-hour period of a day (from 07:00 to 18:00 hs) for the two approaches with priority. Counts are for weekdays with dry pavement and do not include secondary conflicts. . A blank space means that such conflict types are so rare that the observation of any such conflict in the intersection should be taken as abnormal.




43

Table A3. Standard Daily Conflict Counts for Average and Abnormal Level on Signalized Intersections with 4 Approaches

+===================================================================================================================+ | Traffic Volume:10000 to 25000 Vehicles per Day (Total) Abnormal Daily Level | | of Conflict Count | | Type pf Conflict Average Daily Variance 90th 95th | | Conflict Count Percentile Percentile | |-------------------------------------------------------------------------------------------------------------------| | 1.Same Direction – Left Turn | 134.724 10298.3 270.0 340.0 | | 2.Same Direction – Slow Vehicle | 377.938 4928.9 470.0 500.0 | | 3.Same Direction – Lane Change | 7.621 52.8 17.0 22.0 | | 4.Same Direction – Right Turn | 124.476 2445.1 190.0 220.0 | | 5.Opposed Left Turn | 29.057 211.2 49.0 56.0 | | 6.Angle – Left Turn from Left | 0.463 0.466 1.3 1.9 | | 7.Angle – Through from Left (near approach) | 0.289 0.240 - - | | 8.Angle – Right Turn from Left | 0.333 0.188 0.8 1.1 | | 9.Angle – Left Turn from Right | 0.515 0.125 1.0 1.2 | |10.Angle – Through from Right (far approach) | 0.229 0.118 0.7 1.0 | |11.Angle – Right Turn from Right | 3.707 2.839 6.0 7.0 | |12.Opposed Right Turn on Red | 0.094 0.058 - - | | All Same Direction (1 to 4) | 644.760 25338.4 860.0 930.0 | | All Angle – Through (7 + 10) | 0.519 0.215 1.1 1.4 |=================================================================================================================== | Traffic Volume:10000 to 25000 Vehicles per Day (Total) Abnormal Daily Level | | of Conflict Count | | Type of Conflict Average Daily Variance 90th 95th | | Conflict Count Percentile Percentile | |-------------------------------------------------------------------------------------------------------------------| | 1.Same Direction – Left Turn | 83.644 11613.7 265.0 360.0 | | 2.Same Direction – Slow Vehicle | 669.051 23994.7 870.0 940.0 | | 3.Same Direction – Lane Chance | 18.211 160.6 35.0 43.0 | | 4.Same Direction – Right Turn | 218.625 7587.5 470.0 510.0 | | 5.Opposed Left Turn | 22.001 377.7 48.0 60.0 | | 6.Angle – Left Turn from Left | 0.631 0.824 1.7 2.5 | | 7.Angle – Through from Left (near approach) | 0.140 0.135 - - | | 8.Angle – Right Turn from Left | 0.062 0.022 - - | | 9.Angle – Left Turn from Right | 0.417 0.261 1.1 1.4 | |10.Angle – Through from Right (far approach) | 0.290 0.215 - - | |11.Angle – Right Turn from Right | 2.603 2.268 4.6 5.4 | |12.Opposed Right Turn on Red | 0.227 0.124 - - | | All Same Direction (1 to 4) | 989.531 67198.4 1340.0 1460.0 | | All Angle – Through (7 + 10) | 0.430 0.335 1.1 1.5 | +===================================================================================================================+ Notes: . Conflict Counts are the total number of conflicts on the standard 11-hour period of a day (from 07:00 to 18:00 hs) for all the 4 approaches. Counts are for weekdays with dry pavement and do not include secondary conflicts. . A blank space means that such conflict types are so rare that the observation of any such conflict in the intersection should be taken as abnormal.




44

Table A4. Rate of Accidents to Million of Traffic Conflicts (Coefficient of Variation in %) - Weekdays, Dry Pavement

+===================================================================================================================+ | Type of Intersection | INTERSECTION WITH 4 APPROACHES | |-------------------------------------------------------------------------------------------------------------------| | Type of Control | UNSIGNALIZED | SIGNALIZED | |-------------------------------------------------------------------------------------------------------------------| | Total Traffic Volume | 2500 to 10000 | 10000 to 25000 | 10000 to 25000 | more than 25000 | |===================================================================================================================| | | | * | | | | Rear-end Collisions | | 15.0 | 2.66 | 1.43 | | in Same Direction | - | | | | | Conflicts | | ( 66.96% ) | ( 37.16% ) | ( 30.44% ) | | | | | | | |-------------------------------------------------------------------------------------------------------------------| | | | | | | | Collisions with | | 212 | 185 | 671 | | Opposed Left Turn | - | | | | | Conflicts | | ( 43.61% ) | ( 27.10% ) | ( 43.14% ) | | | | | | | |-------------------------------------------------------------------------------------------------------------------| | | | | | | | Angle Collisions | 489 | 735 | - | - | | in Angle Conflicts | | | | | | | ( 20.60% ) | ( 46.82% ) | | | | | | | | | | | | | | | +===================================================================================================================+ Notes: based on 3 years average accident frequency. * only for Same Direction – Left Turn conflicts (M/CE).


Table A.5. Typical Conversion Factors for Estimating Annually Conflict Counts +===================================================================================================================+ | | DRY PAVEMENT | WET PAVEMENT | | Condition |---------------------------------------------------------------------------------------| | | daily correction | days in year | daily correction | days in year | |===================================================================================================================| | | | | | | | WEEKDAYS | 70% | 4/7 | 70% | 1/7 | | | | | | | |-------------------------------------------------------------------------------------------------------------------| | | | | | | | WEEKENDS | 70% | 3/14 | 70% | 1/14 | | | | | | | +===================================================================================================================+ Notes: generic values (specific factors could be obtained for each location, with specific data)




45

Observer: ______________________________ Date: _____________ - Period: _____________ Site: ___________________________________ Approach: ______________________________ Direction: _______________________________ Weather: ________________________________

Sketch of Site:

Actor Codes:

Action Codes:

Time Actor 1 Action Actor 2 Action Comments

Figure A2. Form for Field Observation of Traffic Conflicts in the U.S.ITE Manual (based on ROBERTSON/alli/1994)



46

B. Forms and Criteria of the Swedish Guide on the Traffic Conflict Technique (of the LIT-

Lund Institute of Technology)



47

Traffic Conflict Annotation Form Observer: ________________________ Date: ____________ Time: _______ No. _____ City: ____________________________ Intersection: ____________________________ Weather Condition: ________________ Pavement Condition: _____________________

Sketch

N/S User

I User

II Other Users

Type of User

Sex and Age

Speed (km/h)

Distance (m)

TA value

Action: braking? swerving? running?

Swerve possible?

Feature _______?

Feature _______?

Description of causes for this conflict: ________________________________________ ________________________________________ ________________________________________ ________________________________________

Sketch for this conflict, including:

- your position as and camera position (if used) as

- position and flow direction of each user (at start of the evasive action ...) as for usual vehicles (A=autos, T=trucks, B=buses, O=others) for 2-whellers (b=bicycles, m=motorcycles, o=others) for pedestrians

- identify user I, II or O, as annotated.

Figure B1. Form for Field Observation of Traffic Conflict in the Swedish LIT Guide

(based on ALMQVIST/1998)



48

Figure B2. Criterion for Classifying the Severity of Traffic Conflict in the Swedish LIT Guide (ALMQVIST/1998)



49

THE SWEDISH TRAFFIC CONFLICT TECHNIQUE

Observers Manual General The study of traffic conflicts must be conducted using an observation plan and especially trained observers. When several observers can be used in a study, the observation area of each observer must clearly defined, previously. It is recommended that the person in charge of the study planning and organization is not involved in the data gathering tasks, as observer, to avoid the risk of producing results that are expected by the planner. Also, it is recommended to insert resting intervals between observation periods so as to keep the attention level of the observer (or group of observers).

Preparing the study Every conflict situation is registered in the conflict annotation form, including information on the estimates of speed and distance of the reacting user at the beginning of the evasive action, that will be used to evaluate the time to Accident (TA) value. The distances between notable points must be recognized previously, either through measuring devices or just counts of steps, so as to make easier to estimate the distances involved in each conflict. The estimates of speed should also be trained in the same intersection, before the observation, so as to permit to familiarize the observers to the speeds usually practiced in the study area as measured with portable radars. Observation points When the observer arrives at the study site, it is important to familiarize with the place and check some basic data. Is the intersection to be analyzed correctly identified? What is and how is the observation area to be monitored? After checking the basic data, a good observation point must be selected. The goal is to select a point that offers wide visibility of the study area. For observing traffic conflicts in intersections, the best observation point is usually located at 10 to 25 meters of the intersection. The presence of elevated places near the study area can bring favorable points for better visibility of the site. The observation point should be registered in the annotation form, especially if a before and after study were programmed (so as to repeat the same observation point afterwards). To get valid data on the conflicts, it is essential that the presence of the observer do not change the behavior of the road users. The ideal condition is that the observer and the work apparatus are not visible to the road users or in general, mixing itself to the normal environment of the intersection, as if the observer could be taken as waiting for someone else. It is recommended that the conflict observations were conducted from open space (outside vehicles or buildings), because much useful information or gestual communication between road users can be lost. Also, the comfort and safety of the observers are decisive factors. By this reason, the person in charge of the study planning can use a group of observers for the field work. Comfortable clothes, coffee and



50

water can be brought with the personal appurtenances of each observer for use during the resting intervals. As a rule in field study, the observations should begin always in the scheduled times and the specified duration of the observation should be observed as programmed. To start the annotation in the scheduled time, the observer should arrive to the site at least 15 minutes before the beginning of the observation period. It is important to observe, preliminary, the traffic operation for some minutes to familiarize to the main intersection movements, to the traffic signs and control and to identify any unusual feature of the site. Observation period If the field work is scheduled to last for several hours or several days, the observation process should be conveniently organized, divided in observation periods of 1 hour. During the planning period of the study, the number of days and observation periods per day should be determined as well as the number of annotation forms expected to be used (with a margin) should be prepared. Annotation of conflicts Following the definition, a conflict occurs when 2 road users in a course such that a collision would occur without a chance in speed or path but in which one of the road users is observed to taka an evasive action. Using the estimates of speed and distance, the TA value is obtained from the tabulated values. Then, a conflict is classified as serious if its point in the TA graph (the plot of the pair Speed, TA) is to the up and left sides of the limiting line. If possible, the conflict severity should be checked in the field and only serious conflict should be registered (otherwise, a candidate conflict should be annotated and will be checked afterwards, during the data analysis). Filling the forms The annotation form is filled using the following instructions: The general information (the form header) could be previously filled and checked when some change is noted in the information. For each traffic conflict event one has to register: 1) the hour (day format) 2) the user I and the user II, as involved in the primary conflict. Users should identified and

described in the sketch. If other secondary users are involved in the traffic conflict, the sketch should also identifies and describes them.

3) the speed at the moment when the evasive action is initiated. The speed can be estimated to both road users or only to one of them, the one that undertakes the evasive action.

4) the distance to the imaginary collision point that where the users would meet if keeping the same speed and path as when before starting the evasive action.



51

5) the TA value (Time to the Accident occurrence), obtained from the estimates of speed and distance (using the tabulated values)

6) the evasive action should be described in the way it was undertaken. The common evasive actions are braking or swerving of a vehicle and the stopping and running of a pedestrian.

7) the sketch where the movements and speeds of road users are represented. It is important to detail the sketch to make easy its understanding in the data analysis phase. Other information to identify the study site as geographical references, location of commercial points and businesses activities near the intersection are always important.

8) a description of the sequence of events. In addition to the sketch, the observer must do a short description of the events that occurred and of their severity or intensity. From the sketch and the description, the causes of the conflict event should be clear.

During the annotation period, the observer should take note of traffic violations occurring at the study site, as well as of risky actions, long delays or other relevant factors for the road safety. Preparing for the work Information about the study site location, how to get there, maps of the area, ... should be provided to the observer(s). Required material for field work includes: • a pasta with a set of annotation form for conflict events and the sheet with table/graph for determining the TA value; • some blank sheets for registering additional information, • some pencils, • a clock, • adequate clothes for personal comfort, • personal identification cards, • the phone number of the study supervisor. The observation procedure includes detecting and registering events by estimating speeds and distances, preparing the sketch for the observed event, describing the sequence of actions, trying to identify the causes of the occurring conflicts. Additional information as activities on the study area (vendors, garages) or other events that generate traffic are relevant. The visibility conditions can be an important information as well as the visibility of road signs and traffic signals.



52

C. Forms and Criteria of the French Guide on the Traffic Conflict Technique (of the INRETS-

National Institute for Research on Transport and Safety)



53

TYPES OF CONFLICTS: CLASSIFICATION BASED ON THE ORIGINATING MANEUVER 1. Straight paths: 2. Right turning of vehicle: 3. Left turning of vehicle: 4. Simultaneous turning of vehicles: 5. Pedestrian-vehicle conflicts: 6. Single user conflicts: 61. 62. 65. 69.

11 12 13 15 16

21 22 23 24

31 32 33 34 35

36 37

41 42 43 44 45

51 52 53 54



54

CRITERIA FOR IDENTIFYING THE SEVERITY OF CONFLICTS Light (level 1): the evasive maneuver is clearly commanded for avoiding a collision but the users have the time and distance required for undertaking it without major problems. Despite noting the event, the observer didn’t believe in a real danger of accident.

- for a vehicle or truck, the light conflict is translated by a tight braking but light and initiated at a good distance from the potential collision point, by a swerving in the trajectory or change of queue or lateral displacement effective but at a reasonable distance from the critical point, by a stop that was clearly unpredicted but also without problems, by an increase in the acceleration rate aiming at placing the vehicle ahead of the potential collision point with the other user;

- for a 2-whelled vehicle, the light conflict is related to the same tight braking initiated at a good distance from the critical point, by a lateral displacement not so hard and also controlled, by an unexpected stop with the feet on the ground but without major difficulty, by a sudden acceleration;

- for a pedestrian, the evasive maneuver in the light conflict is less clear and can be a unpredicted stop at the border of the walking side or of the crossing but when the vehicle is approaching yet at a good distance, by walking back, by an increase of the walking speed, all at the same conditions.

Moderate (level 2): the evasive maneuver is clear and can’t be confounded with a normal maneuver and, at the beginning of the evasive action, leaves the doubt on the potential of an accident.

- for a vehicle or a truck, the moderate conflict requires a hard braking but leaves a large distance margin between users at the resolution of the conflict, requires a hard change of queue or lateral displacement but does not leave the users dangerously near, requires a hard acceleration or stop in the same conditions;

- for a 2-whelled vehicle, the moderate conflict also requires a hard braking, hard displacement of change of queue, but yet leaving a distance margin between users at the conflict resolution;

- for a pedestrian, the evasive maneuver in the moderate conflict requires a sudden stop, a jump back or change of path when the vehicle is approaching or stopping but without even a light contact.

Severe (level 3): is the more simple to identify; the evasive maneuver is really hard (emergency or critical) and necessarily precise, yet leaving a small distance between the users at the resolution of the conflict, even a light contact, and the observer believes in the danger of accident up to the last moment.

- for a vehicle or a truck, the severe conflict is translated by a very hard braking with locking wheels of squeezing of tires, even loss of control with sliding, followed by a stop near the other user, by a brutal displacement or hard swerve leaving a small margin for accident avoidance and bringing a real risk of loss of control in the vehicle trajectory, even if temporary, by a very hard acceleration that just permit to cross or overtake by a small distance margin from the conflicting user;

- for a 2-whelled vehicle, the severe conflict also requires a very hard braking that can lock the wheels and bring a loss of control (even of the vertical position of the driver), a very hard displacement that can produce the disequilibrium of the vehicle, both leaving an extremely small margin of distance for avoiding the collision;

- for a pedestrian, the evasive maneuver in the severe conflict is a very quick jump ahead or back (that can’t be taken as a normal behavior of the pedestrian) or a large side jump made due to the light contact with the passing vehicle.



55

Risk Matrix of Injury Accident: 0 → small or null, 1 → medium, 2 → high Type of Conflict Unsignalized Intersection Signalized Intersection

11 1 2 12

0 0

13

0 0

15 1 (or 2 if 2R involved) 2 16

1 (or 2 if 2R involved) 2

21

0 0

22

0 1

23


24


31

0 0

32

0 1

33


34


35


36 0 0 37

0 0

41

0 0

42

0 0

43

0 0

44

0 0

45

0 0

51

1 1

52

0 0

53

1 1

54

0 0

61 to 69

0 0

Note: 2R = 2-wheeled vehicle (motorcycle, bicycle, ...)



56

STANDARD FORM FOR CONFLICT OBSERVATION

STUDY DATA: City: _________________________________ Site: _________________________________________________ Date: _________________________________ Observer: _____________________________________________ Observation Period - Start: _________________________; End: _______________________________________

CONFLICT DATA: Time of Event: Hour [__][__] - Minute [__][__] - Severity of Conflict: [__] Users Involved: Abreviations: VL: auto B: bicicleta

T: taxi M: moto 1 [__] _________________________ Ca: caminhonete Cy: ciclomotor 2 [__] _________________________ PL: caminhão P: pedestre 3 [__] _________________________ Bus: ônibus O: outro 4 [__] _________________________ (distinctive features if present) Sketch of the Event: represent paths, reference number of involved, undertaken maneuvers (braking, stopping, slipping, failure, etc.) Type of Conflict: [__][__] Risk Level: [__][__] Weather: [__] 1 normal 2 chuva Road surface : [__] 1 seca 2 molhada 3 neblina 4 neve 3 neve Temporary problems: _________________________________________________________________________ Comments: __________________________________________________________________________________ ____________________________________________________________________________________________ ____________________________________________________________________________________________ Figure C1. Form for Field Observation of Traffic Conflicts in French Guide (from MUHLRAD/1988).



57

D. Forms and Criteria of the British Guide on the Traffic Conflict Technique (of the TRL-

Transport Research Laboratory)



58

Severity - Classification of Conflicts Conflicts, in addition to involving different maneuvers, vary also by the severity level, determined considering the 4 following factors: A. How long in time, before the potential accident, did the evasive action commence? B. How severe was the evasive action? C. Was the evasive action simple or complex? D. How close did the conflicting vehicles get?

Factor A – Time to collision There are 3 levels of severity for this factor:

• Long; • Moderate; • Short.

At the point when the evasive action begins, i.e., when the vehicle starts to brake of change course, the observer must estimate how long in time it would be before the vehicles collided if this avoidable action was not been undertaken. The observer must remember that in order to make this estimation the DISTANCE BETWEEN, the SPEEDS and the DIRECTIONS of the conflicting vehicles must be taken into account. Consider the example below of one conflict in a T junction.

Obviously the Distance between the vehicles X and Y influences the time to collision, in that the closer the vehicles are at the point when the evasive action begins then the less would be the time before the potential accident between X and Y.

* *

X Y

The Speeds and Distance between vehicles X and Y also have influence, since the higher the speed of X then the shorter the time to collision, but Y´s speed also has to be because Y is travelling away from X. Therefore the higher the speed of Y the longer the time to collision.



59

However X and Y were approaching each other such that a head on collision was imminent, then the higher the speeds of both vehicles the shorter would be the time to collision. It is not possible to give any diagrammatical representation of the 3 levels of factor A.

Factor B – Severity of the evasive action. There are 4 possible intensity (or severity) levels for this factor:

• Light, • Medium, • Heavy, • Emergency.

Again, this will partly depend upon speed and distance between vehicles. The higher the speed of the avoiding vehicle and the smaller the distance between vehicles then the more severe the evasive action is likely to be. Remember there are two types of evasive action possible, Braking and Changing Lane or Course. For braking, the following behaviour is associated with each severity level: - Light: A period of slight controlled braking. - Medium: More prolonged slight controlled braking or a shorter period of sharper controlled

braking in which the front of the vehicle would be seen to dip down. - Heavy: Prolonged sharper, less controlled braking where the front of the vehicle dips

abruptly. This “heavy” braking may involve some squealing of types. - Emergency: Uncontrolled, very heavy, continuous braking where the wheels may lock up

and the vehicle skip out of control. For change of lane or course, the following behaviour is associated with each severity level: - Light: A controlled, slight change of course. - Medium: A controlled, complete change of lane. - Heavy: A less controlled, sudden swerve to change course. - Emergency: Very heavy, uncontrolled swerving.



60

By virtue of the nature of the factor, it is difficult to infer the differences between the severity levels from static diagrams.

Factor C – Complexity of evasive action. There are 2 alternatives for this factor:

• Simple, • Complex.

- Simple: A single action EITHER braking OR a change of course only occurs. - Complex: More than one action, braking AND change of course must occur.

Examples are shown below to illustrate simple evasive action and also a complex action: Simple Simple Complex

* * X Y

X Y

* *

X Y

Factor D – Proximity of conflicting vehicles. There are 3 possibility differentiated from each other on the basis of the distance between the conflicting vehicles at the point where the evasive action is terminated, and/or they are closest and still on a collision course. The 3 categories are:

! more than 2 car lengths (> 2 vehicles), ! 1 to 2 car lengths (1-2 vehicles), ! less than 1 car length (<1 vehicle).



61

> 2 vehicles 1-2 vehicles <1 vehicle

More than 30 ft Between 15 and 30 ft Below 15 ft approx. approx. approx. The final severity grade of a conflict is obtained by refering to Table 4, which defines the relationship between the different combinations of the various levels of the 4 factors, A-D, and the different conflict grades.



62

FACTORS GRADES

FACTOR A Time to

Collision

FACTOR B Severity of

Evasive Action

FACTOR C Complexity of Evasive Action

FACTOR D Proximity of

Conflicting Vehicles

1.

Long Long Long Moderate Moderate Moderate Short

Light Medium Light Light Light Medium Medium

Simple/Complex Simple/Complex Simple/Complex Simple/Complex Simple Simple/Complex Simple/Complex

>2 vehicles >2 vehicles 1 - 2 vehicles > 2 vehicles 1 - 2 vehicles > 2 vehicles > 2 vehicles

2.

Long Long Long Moderate Moderate Moderate Moderate Short Short Short Short

Light Medium Heavy Light Light Medium Heavy Light Light Medium Heavy

Simple/Complex Simple/Complex Complex Simple/Complex Simple/Complex Simple/Complex Complex Simple/Complex Simple/Complex Simple/Complex Simple/Complex

< 1vehicle 1 - 2 vehicles > 2 vehicles < 1 vehicle 1 - 2 vehicles 1 - 2 vehicles > 2 vehicles 1 - 2 vehicles < 1 vehicle 1 - 2 vehicles > 2 vehicles

3.

Long Long Moderate Moderate Moderate Moderate Short Short Short Short Short

Medium Heavy Light Medium Heavy Heavy Medium Medium Heavy Emergency Emergency

Simple/Complex Simple/Complex Complex Simple/Complex Simple/Complex Simple/Complex Complex Simple/Complex Simple/Complex Simple/Complex Simple/Complex

< 1 vehicle 1 - 2 vehicles < 1 vehicle < 1 vehicle 1 - 2 vehicles < 1 vehicle 1 - 2 vehicles < 1 vehicle 1 - 2 vehicles > 2 vehicles 1 - 2 vehicles

4.

Moderate Short Short

Emergency Heavy Emergency

Simple/Complex Simple/Complex Simple/Complex

< 1 vehicle < 1 vehicle < 1 vehicle

Table D1 – Conversion of Factor Levels to Conflict Grades (BAGULEY/1988).



63

Figure D1. Form for Field Observation of Traffic Conflicts in British Guide (based on BAGULEY/1988).

Site: ________________________________ Day: ________________________________ Date: _________________________________ Observer: ___________________________ Weather: ____________________ _____________________________________

Vehicle 1: is the one whose action cause the conflict Vehicle 2: is the one that has to take the avoiding action

Time of Conflict

Vehicle 1 Type:

C=Car B=Bus L=Light GV H=Heavy GV M=Motorcycle

Vehicle 1 Maneuver

Vehicle 2 Type:

C=Car B=Bus L=Light GV H=Heavy GV M=Motorcycle

Vehicle 2 Maneuver

Time to Collision L = Long M = Moderate S = Short

Severity of Evasive Action L=Light M=Medium H=Heavy E=Emergency

Complexity of Evasive Action S = Simple C = Complex

Proximity of Conflicting Vehicles < 1 vehicle 1–2 vehicles > 2 vehicles

Comments Causes of Conflict

LEMT-Laboratório de Estudos Metodológicos em Tráfego e Transportes

Land Use and Transport Integrated Models

i

SUMMARY

1. Introduction 1

2. The Identification, Classification and Analysis of Pedestrian-Vehicle Traffic Conflicts 3

3. Methods for Determining and Evaluating Diagnostic Parameters for Pedestrian-Vehicle Conflict Studies 13

4. Application to the Determination of Diagnostic Parameters for Pedestrian-Vehicle Conflicts in the São Paulo Study 21

5. Conclusions 34

References 36

Appendices 38

A. Forms and Criteria of the U.S.Guide on the Traffic Conflict Technique (from the FHWA-Federal Highway Administration and the ITE-Institute of Traffic Engineers)

B. Forms and Criteria of the Swedish Guide on the Traffic Conflict Technique (from the LIT-Lund Institute of Technology)

C. Forms and Criteria of the French Guide on the Traffic Conflict Technique (from the INRETS-National Institute for Research on Transport and Safety)

D. Forms and Criteria of the British Guide on the Traffic Conflict Technique (from the TRL-Transport Research Laboratory)

relatÓrio de trabalho no.06/04 -...

Documents