Three questions are addressed: How many reportable accidents are in fact reported? What is the re... more Three questions are addressed: How many reportable accidents are in fact reported? What is the relationship among road safety, accidents occurring, and accidents reported? What are the effects of incomplete reporting on the estimation of road safety? A review of 18 studies made at different times and localities reveals considerable variability in the degree of nonreporting. As a ballpark estimate, fatalities may be known to an accuracy of about 5 percent. Some 20 percent of injuries requiring hospitalization and perhaps 50 percent of all injuries are not found in official statistics. Furthermore, the probability of an accident's being reported depends on the age of the victim; whether the victim is the driver, the passenger, or a nonoccupant; the number of vehicles involved; and several additional factors. Analysis shows that the accuracy with which road safety can be measured depends on the proportion of accidents reported and on the accuracy with which this proportion is known. It appears that the variance of the estimate of safety (or of the safety effect of some measure) is inversely proportional to the square of the average proportion of accidents reported.
Suppose that there was a ‘well specified’ regression model, one in which the correct predictor va... more Suppose that there was a ‘well specified’ regression model, one in which the correct predictor variables were combined into the correct function, and that the unknown parameters were estimated using good and plentiful data. Can such a model be used to predict what change in the response variable is caused by a change in one of the predictor variables? Surprisingly the answer is: “No.” In this paper the author identifies the condition that often frustrates the causal interpretation of well specified regressions. The author shows how this very condition led astray several authors who used regressions to estimate the role of speed in accident generation
Nothing is known about unreported crashes; the information we have is of reported crashes only. W... more Nothing is known about unreported crashes; the information we have is of reported crashes only. Whether a crash gets reported depends on its severity. It follows by logic that, using only data about reported crashes, it is impossible to say whether a change or difference in crash counts reflects a change or difference in crash frequency or in crash severity. This indeterminacy has practical implications. Examples discussed are of the misattribution of overrepresentation in reported crashes of older drivers and of trucks to causal factors related to the frequency of crash involvement, and the misinterpretation by researchers of findings about the roll-over propensity of SUVs.
The design of traffic management schemes usually requires knowledge of the pattern of trips on th... more The design of traffic management schemes usually requires knowledge of the pattern of trips on the system under scrutiny. This pattern is ordinarily described by an origin-destination (O-D) flow matrix. One common task of this type of matrix is the estimation of flows between the intersection approaches on a stretch of road. Estimation is based on intersection flow counts that are supplemented by a license-plate survey. In this paper a procedure is developed to obtain the most likely O-D flow estimates by using both intersection counts and results of the license-plate survey. The procedure is described in detail on the basis of a numerical example. An earlier paper reported a method of estimation that relies on intersection counts only and does not require the conduct of a sample license-plate survey. An empirical examination is conducted to test how estimation accuracy increases when the added information from the license-plate survey is used. This examination reveals that when the...
To manage unsafety we need to know how much safer is "A" then "B". Unfortunat... more To manage unsafety we need to know how much safer is "A" then "B". Unfortunately, in spite of extensive experience with many measures, too often their impact on unsafety is not known. It is truly difficult estimate differences in unsafety. To answer the question of "how much safer" one has to guess at "what would have been". What method of guessing is good is an empirical question. In this light I examine the problem of choosing a comparison system. Next I argue that the question is not about whether a null hypothesis can be rejected but about what is the magnitude of the effect on unsafety of some intervention. The harm of repeated significance testing is illustrated in the case of the right turn on red controversy. Another cause of ignorance is that the management of road safety is the responsibility of public bodies. People who lead and served these bodies seldom find it in their interest to find out what effect on unsafety their programs h...
Papers presented at this session include: preliminary location of passing lanes using a simulatio... more Papers presented at this session include: preliminary location of passing lanes using a simulation model (morrall,j); safety effect of resurfacing rural highways (sabo,pa and hauer,e). For the covering abstract of the conference see IRRD 291074. (TRRL)
Most agencies with responsibility for extensive road systems use some variant of the rate-and-num... more Most agencies with responsibility for extensive road systems use some variant of the rate-and-number method to identify hazardous locations or blackspots. Sites so identified are later examined in detail to diagnose deficiencies and to suggest remedial measures. In this paper the degree to which the rate-and-number method is successful in identifying the unidentified, and what proportion of the sites that are subjected to detailed examination are not deviant at all, is examined. The first part of the paper is devoted to the analysis and development of the mathematical machinery. In the second part the use of the analytical results is illustrated by application to two data sets--one dealing with highway ramps in Ontario and the other with California drivers. The main result of this research is the facility to examine the performance of various identification procedures on the basis of measures of performance that are easy to understand. Such an examination should lead to a realistic ...
In earlier work, the authors developed accident prediction models for the general population of O... more In earlier work, the authors developed accident prediction models for the general population of Ontario drivers based on age, gender, conviction record and accident history. This paper presents the results of an extension of that work to the separate populations of truck and car drivers. For each sub-population, a set of linear regression models was calibrated using different combinations of variables depending on the weightings and groupings assigned to convictions and accidents, and whether or not age and gender were included as variables. The models were used to predict which drivers were likely to have accidents in a subsequent two-year period based on their driving record in an initial two-year period.
The current demerit point system in use in Ontario allocates points to offences on the basis of t... more The current demerit point system in use in Ontario allocates points to offences on the basis of the perceived seriousness of the offence. The goal of the work reported here was to allocate points to offences with a different purpose in mind. The purpose here was to use a driver's record of convictions and accidents in order to predict, as well as possible, which drivers, based on their past record, are most likely to have an accident in the near future. A sample of 827,955 records of Ontario drivers containing information about age, gender, convictions, accidents, demerit points and suspensions, for 1981-1984 was examined. On this basis 16 new models to estimate a driver's accident potential were formulated. Some models used age and gender information, others did not. Some models assigned different weights to each of the conviction categories; in other models all convictions were weighted equally (equivalent to 1 point per conviction). In some models accident data were used;...
Data on accidents, road characteristics, and traffic for rural, two-lane roads in seven states ha... more Data on accidents, road characteristics, and traffic for rural, two-lane roads in seven states have been assembled. It was found that, for the same amount of traffic, different states record widely discrepant numbers of accidents. The discrepancy does not disappear even when roads with the same lane width, shoulder type, and terrain are examined. It is concluded that (a) data from different states should not be pooled, (b) warrants and standards based on accidents should be tailored to each state, and (c) the cause of the noted differences should be investigated.
Abstract: In the early 1980s, two kinds of resurfacing projects were undertaken in New York State... more Abstract: In the early 1980s, two kinds of resurfacing projects were undertaken in New York State: Fast track projects involving only resurfacing, and reconditioning and preservation (R&P) projects in which roadside and roadway safety improvements have been ...
This report documents and presents the results of a study to develop an annotated outline, protot... more This report documents and presents the results of a study to develop an annotated outline, prototype chapter, and work plan for the first edition of the Highway Safety Manual (HSM). The purpose of the HSM will be to provide the best factual information and tools, in a useful and widely accepted form, to facilitate roadway planning, design, and operational decisions based upon explicit consideration of their safety consequences.
The Transportation Research Board (TRB) and the American Association of State Highway and Transpo... more The Transportation Research Board (TRB) and the American Association of State Highway and Transportation Officials (AASHTO) have begun a major initiative to develop a Highway Safety Manual (HSM). The HSM began from recognition of the fact that, for safety to receive proper consideration in the highway project development process, analysts need tools to make quantitative statements about the safety effects of proposed projects or design alternatives. The key components of the HSM will be: Part I—Introduction and Fundamentals; Part II—Knowledge; Part III—Predictive Methods; Part IV—Safety Management of a Roadway System; and Part V—Safety Evaluation. HSM Part III will consist of chapters that provide safety prediction methodologies for specific types of highway facilities. The TRB Task Force plans that the first edition of the HSM should include safety prediction methodologies for rural two-lane highways, rural multilane highways, and urban and suburban arterials. These methodologies w...
A sample of 827,955 records of Ontario drivers containing information about age, gender, convicti... more A sample of 827,955 records of Ontario drivers containing information about age, gender, convictions, accidents, demerit points, and suspensions for 1981-1984 has been examined. On this basis 16 alternate models to estimate a driver's accident potential have been formulated. It appears that the currently used demerit point system, wherein the number of points associated with an offense reflects the perceived seriousness of the offense, is not a good predictor of accident potential. One can predict better by relying on the driver's record of accidents and convictions and still better by making use of a model for which the "regression weights" have been rigorously estimated. The performance of alternative models for the estimation of drivers' accident potential is described in terms of "hits" and "false alarms." It is shown, for example, that if the top 10,000 drivers are selected by the best model, 3,757 of these are expected to have an accid...
The enforcement of traffic laws is based on the belief that it induces greater driver compliance ... more The enforcement of traffic laws is based on the belief that it induces greater driver compliance with the rules of the road, which leads to a reduction in the number of accidents. Conclusive empirical evidence substantiating this belief has, however, not previously been presented. The availability of computerized accident records for the Metropolitan Toronto Police Selective Enforcement program provided a unique opportunity to test whether increased enforcement was followed by a reduction in the number of accidents. By using accident records for 1800 locations over a period of 4 years, estimates of accident rates were obtained that accounted for a time trend and seasonal variations. The expected number of accidents so obtained was compared to the number of recorded accidents. Locations that received increased enforcement showed consistently fewer than the expected number of accidents. In the experiment, all important factors except increase in enforcement are randomized. Thus, unles...
Transportation Research Record: Journal of the Transportation Research Board, 2004
The hope is that statistical models fitted to historical data can be used to estimate the effect ... more The hope is that statistical models fitted to historical data can be used to estimate the effect of road design elements on safety. Whether this can be done is not clear. A sign of trouble is that models based on diverse data sets tend not to yield similar results. Suggestions are made on how to increase the chance of success in this quest. Emphasis is on three questions: Which variables should serve in the model? What mathematical function should represent their influence? How does one check whether the representation of the influence of a variable is appropriate?
The essence of CMFs is that they are to capture the safety effect of causes. The commonly used &a... more The essence of CMFs is that they are to capture the safety effect of causes. The commonly used 'before-and-after' and 'regression modeling' approaches to CMF estimation often run into problems and consensus about trustworthy CMF estimates is slow to emerge. The nature of these problems and some promising directions to alleviate them are examined in the companion paper . In view of the identified directions, what alternative approaches to the determination of the safety effect of causes hold promise? This is the question to be examined here.
Three questions are addressed: How many reportable accidents are in fact reported? What is the re... more Three questions are addressed: How many reportable accidents are in fact reported? What is the relationship among road safety, accidents occurring, and accidents reported? What are the effects of incomplete reporting on the estimation of road safety? A review of 18 studies made at different times and localities reveals considerable variability in the degree of nonreporting. As a ballpark estimate, fatalities may be known to an accuracy of about 5 percent. Some 20 percent of injuries requiring hospitalization and perhaps 50 percent of all injuries are not found in official statistics. Furthermore, the probability of an accident's being reported depends on the age of the victim; whether the victim is the driver, the passenger, or a nonoccupant; the number of vehicles involved; and several additional factors. Analysis shows that the accuracy with which road safety can be measured depends on the proportion of accidents reported and on the accuracy with which this proportion is known. It appears that the variance of the estimate of safety (or of the safety effect of some measure) is inversely proportional to the square of the average proportion of accidents reported.
Suppose that there was a ‘well specified’ regression model, one in which the correct predictor va... more Suppose that there was a ‘well specified’ regression model, one in which the correct predictor variables were combined into the correct function, and that the unknown parameters were estimated using good and plentiful data. Can such a model be used to predict what change in the response variable is caused by a change in one of the predictor variables? Surprisingly the answer is: “No.” In this paper the author identifies the condition that often frustrates the causal interpretation of well specified regressions. The author shows how this very condition led astray several authors who used regressions to estimate the role of speed in accident generation
Nothing is known about unreported crashes; the information we have is of reported crashes only. W... more Nothing is known about unreported crashes; the information we have is of reported crashes only. Whether a crash gets reported depends on its severity. It follows by logic that, using only data about reported crashes, it is impossible to say whether a change or difference in crash counts reflects a change or difference in crash frequency or in crash severity. This indeterminacy has practical implications. Examples discussed are of the misattribution of overrepresentation in reported crashes of older drivers and of trucks to causal factors related to the frequency of crash involvement, and the misinterpretation by researchers of findings about the roll-over propensity of SUVs.
The design of traffic management schemes usually requires knowledge of the pattern of trips on th... more The design of traffic management schemes usually requires knowledge of the pattern of trips on the system under scrutiny. This pattern is ordinarily described by an origin-destination (O-D) flow matrix. One common task of this type of matrix is the estimation of flows between the intersection approaches on a stretch of road. Estimation is based on intersection flow counts that are supplemented by a license-plate survey. In this paper a procedure is developed to obtain the most likely O-D flow estimates by using both intersection counts and results of the license-plate survey. The procedure is described in detail on the basis of a numerical example. An earlier paper reported a method of estimation that relies on intersection counts only and does not require the conduct of a sample license-plate survey. An empirical examination is conducted to test how estimation accuracy increases when the added information from the license-plate survey is used. This examination reveals that when the...
To manage unsafety we need to know how much safer is "A" then "B". Unfortunat... more To manage unsafety we need to know how much safer is "A" then "B". Unfortunately, in spite of extensive experience with many measures, too often their impact on unsafety is not known. It is truly difficult estimate differences in unsafety. To answer the question of "how much safer" one has to guess at "what would have been". What method of guessing is good is an empirical question. In this light I examine the problem of choosing a comparison system. Next I argue that the question is not about whether a null hypothesis can be rejected but about what is the magnitude of the effect on unsafety of some intervention. The harm of repeated significance testing is illustrated in the case of the right turn on red controversy. Another cause of ignorance is that the management of road safety is the responsibility of public bodies. People who lead and served these bodies seldom find it in their interest to find out what effect on unsafety their programs h...
Papers presented at this session include: preliminary location of passing lanes using a simulatio... more Papers presented at this session include: preliminary location of passing lanes using a simulation model (morrall,j); safety effect of resurfacing rural highways (sabo,pa and hauer,e). For the covering abstract of the conference see IRRD 291074. (TRRL)
Most agencies with responsibility for extensive road systems use some variant of the rate-and-num... more Most agencies with responsibility for extensive road systems use some variant of the rate-and-number method to identify hazardous locations or blackspots. Sites so identified are later examined in detail to diagnose deficiencies and to suggest remedial measures. In this paper the degree to which the rate-and-number method is successful in identifying the unidentified, and what proportion of the sites that are subjected to detailed examination are not deviant at all, is examined. The first part of the paper is devoted to the analysis and development of the mathematical machinery. In the second part the use of the analytical results is illustrated by application to two data sets--one dealing with highway ramps in Ontario and the other with California drivers. The main result of this research is the facility to examine the performance of various identification procedures on the basis of measures of performance that are easy to understand. Such an examination should lead to a realistic ...
In earlier work, the authors developed accident prediction models for the general population of O... more In earlier work, the authors developed accident prediction models for the general population of Ontario drivers based on age, gender, conviction record and accident history. This paper presents the results of an extension of that work to the separate populations of truck and car drivers. For each sub-population, a set of linear regression models was calibrated using different combinations of variables depending on the weightings and groupings assigned to convictions and accidents, and whether or not age and gender were included as variables. The models were used to predict which drivers were likely to have accidents in a subsequent two-year period based on their driving record in an initial two-year period.
The current demerit point system in use in Ontario allocates points to offences on the basis of t... more The current demerit point system in use in Ontario allocates points to offences on the basis of the perceived seriousness of the offence. The goal of the work reported here was to allocate points to offences with a different purpose in mind. The purpose here was to use a driver's record of convictions and accidents in order to predict, as well as possible, which drivers, based on their past record, are most likely to have an accident in the near future. A sample of 827,955 records of Ontario drivers containing information about age, gender, convictions, accidents, demerit points and suspensions, for 1981-1984 was examined. On this basis 16 new models to estimate a driver's accident potential were formulated. Some models used age and gender information, others did not. Some models assigned different weights to each of the conviction categories; in other models all convictions were weighted equally (equivalent to 1 point per conviction). In some models accident data were used;...
Data on accidents, road characteristics, and traffic for rural, two-lane roads in seven states ha... more Data on accidents, road characteristics, and traffic for rural, two-lane roads in seven states have been assembled. It was found that, for the same amount of traffic, different states record widely discrepant numbers of accidents. The discrepancy does not disappear even when roads with the same lane width, shoulder type, and terrain are examined. It is concluded that (a) data from different states should not be pooled, (b) warrants and standards based on accidents should be tailored to each state, and (c) the cause of the noted differences should be investigated.
Abstract: In the early 1980s, two kinds of resurfacing projects were undertaken in New York State... more Abstract: In the early 1980s, two kinds of resurfacing projects were undertaken in New York State: Fast track projects involving only resurfacing, and reconditioning and preservation (R&P) projects in which roadside and roadway safety improvements have been ...
This report documents and presents the results of a study to develop an annotated outline, protot... more This report documents and presents the results of a study to develop an annotated outline, prototype chapter, and work plan for the first edition of the Highway Safety Manual (HSM). The purpose of the HSM will be to provide the best factual information and tools, in a useful and widely accepted form, to facilitate roadway planning, design, and operational decisions based upon explicit consideration of their safety consequences.
The Transportation Research Board (TRB) and the American Association of State Highway and Transpo... more The Transportation Research Board (TRB) and the American Association of State Highway and Transportation Officials (AASHTO) have begun a major initiative to develop a Highway Safety Manual (HSM). The HSM began from recognition of the fact that, for safety to receive proper consideration in the highway project development process, analysts need tools to make quantitative statements about the safety effects of proposed projects or design alternatives. The key components of the HSM will be: Part I—Introduction and Fundamentals; Part II—Knowledge; Part III—Predictive Methods; Part IV—Safety Management of a Roadway System; and Part V—Safety Evaluation. HSM Part III will consist of chapters that provide safety prediction methodologies for specific types of highway facilities. The TRB Task Force plans that the first edition of the HSM should include safety prediction methodologies for rural two-lane highways, rural multilane highways, and urban and suburban arterials. These methodologies w...
A sample of 827,955 records of Ontario drivers containing information about age, gender, convicti... more A sample of 827,955 records of Ontario drivers containing information about age, gender, convictions, accidents, demerit points, and suspensions for 1981-1984 has been examined. On this basis 16 alternate models to estimate a driver's accident potential have been formulated. It appears that the currently used demerit point system, wherein the number of points associated with an offense reflects the perceived seriousness of the offense, is not a good predictor of accident potential. One can predict better by relying on the driver's record of accidents and convictions and still better by making use of a model for which the "regression weights" have been rigorously estimated. The performance of alternative models for the estimation of drivers' accident potential is described in terms of "hits" and "false alarms." It is shown, for example, that if the top 10,000 drivers are selected by the best model, 3,757 of these are expected to have an accid...
The enforcement of traffic laws is based on the belief that it induces greater driver compliance ... more The enforcement of traffic laws is based on the belief that it induces greater driver compliance with the rules of the road, which leads to a reduction in the number of accidents. Conclusive empirical evidence substantiating this belief has, however, not previously been presented. The availability of computerized accident records for the Metropolitan Toronto Police Selective Enforcement program provided a unique opportunity to test whether increased enforcement was followed by a reduction in the number of accidents. By using accident records for 1800 locations over a period of 4 years, estimates of accident rates were obtained that accounted for a time trend and seasonal variations. The expected number of accidents so obtained was compared to the number of recorded accidents. Locations that received increased enforcement showed consistently fewer than the expected number of accidents. In the experiment, all important factors except increase in enforcement are randomized. Thus, unles...
Transportation Research Record: Journal of the Transportation Research Board, 2004
The hope is that statistical models fitted to historical data can be used to estimate the effect ... more The hope is that statistical models fitted to historical data can be used to estimate the effect of road design elements on safety. Whether this can be done is not clear. A sign of trouble is that models based on diverse data sets tend not to yield similar results. Suggestions are made on how to increase the chance of success in this quest. Emphasis is on three questions: Which variables should serve in the model? What mathematical function should represent their influence? How does one check whether the representation of the influence of a variable is appropriate?
The essence of CMFs is that they are to capture the safety effect of causes. The commonly used &a... more The essence of CMFs is that they are to capture the safety effect of causes. The commonly used 'before-and-after' and 'regression modeling' approaches to CMF estimation often run into problems and consensus about trustworthy CMF estimates is slow to emerge. The nature of these problems and some promising directions to alleviate them are examined in the companion paper . In view of the identified directions, what alternative approaches to the determination of the safety effect of causes hold promise? This is the question to be examined here.
Uploads
Papers by Ezra Hauer