Road Safety Engineering BCV755A

Module – 2
Traffic Engineering Studies
Structure
2.0 Introduction
2.1 Objectives
2.2 Statistical Methods in Traffic Safety Analysis
2.3 Traffic Management Measures
2.4 Recommended questions
2.5 Outcomes
2.6 Further Reading

2.0 Introduction
The spectacular increase in the number of motor vehicles on the road has created a
major social problem-the loss of lives through road accidents. The appalling human misery
and the serious economic loss caused by road accidents demand the attention of the society
and call for the solution of the problem. A multi-disciplinary approach is needed in
understanding the problem and providing solutions. The traffic engineer is concerned because
many features of the highway affect the safety of the vehicle and the other road users. The
analysis of the accident statistics provides clues to the many factors that lead to the accidents
and to improvements that may be desired. Based on the statistics the traffic engineer must
devise ways to reduce the accidents through better planning, design, construction,
maintenance and traffic operation. The traffic engineer is also concerned about the regulation
and management of traffic to ensure safer travel. Accident data supply valuable information
to control, regulate and manage the traffic more efficiently. The cost of traffic accidents helps
the traffic engineer in evaluating an improvement scheme aimed at reducing the accidents

Need
In ordinary sense, Statistics deals with the collection and compilation of data. The
subject of Statistics has outgrown this simple definition and is now more and more concerned
about drawing conclusions, and inferences from the observed data. Statistical methods find
important applications in many fields of science and engineering. Traffic engineering is a
subject which deals extensively with quantitative data in the planning, design and operation
of transportation facilities. Statistical methods are a powerful tool to analyse and interpret
these data. In this Chapter, the elementary principles of Statistics are outlined, with particular
reference to their application to traffic engineering studies.

Purpose of Traffic Studies

 Establish relative importance of any road
 Planning, designing and regulation of traffic
 Structural design of road
 Planning and design of new streets
 Establishment properties and schedules for traffic improvements.
 Design of intersection and other control devices

Department of Civil Engineering, ATMECE, Mysuru Page 1

Road Safety Engineering BCV755A

2.1 Objectives
 To provide students with a comprehensive understanding of the principles, strategies,
and techniques related to ensuring safety on roadways

2.2 Statistical Methods in Traffic Safety Analysis

A number of statistical methods are currently being applied in traffic engineering
studies. These include:
1. Regression method
2. Poisson distribution
3. Chi- Squared Distribution
4. Statistical Comparison

2.2.1 Regression Methods

The basic principle behind this method is that the expected number of accidents, on a
certain road system during a given time period, is dependent in a linear way on factors which
are supposed to be of significance for the determination of accident frequency
The number of accidents occurring on a certain day is itself assumed to be normally
distributed, with a mean value being a linear function of regression variables, and a variance
being constant a same for all days of a certain time period studied.
The numbers of accidents on different days are assumed to be stochastically
independent.

Some of the regression variables (independent variables) that could be considered are:
 Two wheeled vehicles involved in personal injury accidents as a proportion of all
vehicles involved.
 Cost of safety improvements.
 Number of pedestrians
 Pavement width
 Number of junctions per km length of road

Use of Regression Analysis in Traffic Safety Studies: In traffic safety, regression models
are commonly applied to develop Safety Performance Functions (SPFs), which estimate the
expected number of crashes at a location based on its characteristics. These functions are
used to identify high-risk areas, evaluate countermeasures, and perform cost-benefit analyses.
A typical regression model in this context might relate accident frequency (dependent
variable) to one or more independent variables such as:
 Traffic volume (e.g., Average Daily Traffic - ADT)
 Roadway design features (e.g., number of lanes, shoulder width)
 Intersection type (e.g., signalized, unsignalized)
 Environmental factors (e.g., weather, lighting conditions)
Different types of regression models are used depending on the nature of the data.
Since crash data are usually count-based and often over-dispersed (variance greater than the
mean), Poisson regression or Negative Binomial regression is preferred over standard linear

Department of Civil Engineering, ATMECE, Mysuru Page 2

Road Safety Engineering BCV755A

regression. These models are well-suited for predicting the frequency of rare events like road
crashes.

Steps in Applying Regression Analysis in Traffic Safety:

 Data Collection: Collect historical crash data along with relevant variables for road
segments or intersections.
 Model Specification: Choose an appropriate regression model (Poisson, Negative
Binomial, or Zero-Inflated models) depending on the distribution of crash data.
 Parameter Estimation: Use statistical software to estimate the coefficients that show
the strength and direction of the relationship between each independent variable and
crash frequency.
 Model Validation: Evaluate the model’s accuracy using statistical tests (e.g.,
goodness-of-fit, likelihood ratio tests) and by comparing predicted crash values with
actual crash data.
 Prediction and Analysis: Apply the model to new locations to estimate expected
crash frequencies, which can then inform safety prioritization and intervention
planning.

Benefits of Using Regression Models for Predicting Accident Frequency:

 Data-Driven Decision-Making: Regression models provide objective, quantitative
insights that help identify high-risk locations based on expected crash frequencies
rather than relying only on historical crash counts.
 Control for Exposure: By including traffic volume as an independent variable,
regression models adjust for differences in exposure, allowing for fair comparisons
between locations with varying traffic levels.
 Identification of Contributing Factors: These models help isolate and quantify the
effect of specific roadway or environmental features on crash likelihood, supporting
targeted improvements.
 Prediction for Untreated Sites: Regression models allow prediction of crash
frequency at locations without prior crash history, enabling proactive safety
management.
 Support for Cost-Benefit Analysis: By estimating crash reduction potential,
regression models help assess the economic viability of proposed safety
countermeasures.

Regression analysis is a cornerstone of modern traffic safety analysis, offering a

statistically sound method to predict accident frequency and identify influencing factors. By
using regression models, transportation agencies can prioritize interventions more effectively,
implement evidence-based safety strategies, and ultimately reduce traffic crashes and save
lives.
Multiple Linear Regression
A statistical technique which will be most frequently encountered by a Traffic
Engineer and Transport Planner is the multiple linear regression analysis. The problem

Department of Civil Engineering, ATMECE, Mysuru Page 3

Road Safety Engineering BCV755A

concerns with the establishment of relationship between a variable which is known to

respond to changes in two or more other variables. The variable which is known to respond,
Y variable, is commonly called the dependent variable, and the other variables influencing it
are called the independent variables, i.e., X variables. The function will be of the following
form:
𝑌 = 𝑎 + 𝑎 𝑥 + 𝑎 𝑥 + ⋯+ 𝑎 𝑥
Where
Y = True estimate of the dependent variable, y
𝑎1, 𝑎2 ... 𝑎n = Regression coefficients of the representative m independent variables
x1, x2 … x3 = m independent variables
𝑎0 = Regression constant

Assumptions in the multiple linear regression analysis: The following are some of the
conditions which must be satisfied if a multiple linear regression analysis is to be used:
1. All the independent variables must be independent of each other and there should be
no correlation between them.
2. All the variables are normally distributed.
3. All the variables are continuous.
4. A linear relationship exists between the dependent variable and the independent
variables.
5. The influence of the independent variable is additive, i.e., the inclusion of each
variable in the equation contributes a distinct portion in the estimation of the
dependent variable and all the variables together contribute additively in the
estimation.

2.2.2 Poisson Distribution

The Poisson distribution is a discrete probability distribution that expresses the
probability of a given number of events (such as traffic accidents) occurring in a fixed
interval of time or space, assuming these events happen independently and with a constant
average rate. In the context of traffic safety, the Poisson distribution is widely used to model
accident occurrences at specific locations, such as intersections, road segments, or highways,
over defined time periods.
The Poisson Distribution helps us understand the probability of a certain number of
events happening within a fixed period or area. It is often used to model things like the arrival
of vehicles.
The formula is:
𝒏𝒑𝒓 𝒆 𝒏𝒑 𝒎𝒓 𝒆 𝒎
𝐏(𝐫) = =
𝒓! 𝒓!
where:
P(r) = Probability of a occurrence of an event r times in n trials.
p = Probability of occurrence of the event in a single trial.
e = base of natural logarithms.
m = np

Department of Civil Engineering, ATMECE, Mysuru Page 4

Road Safety Engineering BCV755A

Poisson distribution is normally valid when:

(i) n is sufficiently large, say >> 50 , and
(ii) p is small, say << 0.1 .
In the above expression, it may be noted that
np = m = average number of times the event occurs

Applying the above formula to determine the probability of a driver causing an accident, let:
N be the number of drivers
M be the kilometres driven by each driver
p be the probability of having an accident per kilometre travelled
m be the average rate of occurrence of accidents in a length of travel of M kilometres (m =
p*M)

Then the probability of a driver having r accidents,

𝒎𝒓 𝒆 𝒎
𝐏(𝐫) =
𝒓!
Also, the number of drivers having r accidents
𝒎𝒓 𝒆 𝒎
𝐏(𝐫) = 𝑵
𝒓!
In the above formula, the value of p is very small and the value M is very large.

Application of the Poisson Distribution in Traffic Accident Modeling: In traffic accident

studies, the primary objective is often to model the frequency of crashes and identify factors
that influence crash risk. Since traffic accidents are random, rare, and discrete events, the
Poisson distribution provides a suitable framework for statistical modeling.

Key Assumptions of the Poisson Model:

 Events are independent – One accident does not influence the probability of another.
 The average rate is constant – The mean number of accidents over a time or space
interval remains stable.
 Only one event can occur at a specific point in time or space – Two accidents

Why the Poisson Distribution is Commonly Used in Accident Studies:

 Modeling of Rare Events: Traffic accidents are generally infrequent at any specific
location. The Poisson distribution is particularly well-suited for modeling such rare
and count-based events.
 Simplicity and Interpretability: The model is mathematically straightforward and
easy to implement using statistical software. The single parameter λ (mean crash rate)
is intuitive and directly related to crash frequency.
 Foundation for Advanced Models: While the basic Poisson model assumes equal
mean and variance, real-world crash data often show overdispersion (variance >
mean). In such cases, the Poisson distribution forms the basis for more advanced

Department of Civil Engineering, ATMECE, Mysuru Page 5

Road Safety Engineering BCV755A

models like the Negative Binomial distribution, which introduces an extra parameter
to account for variability.
 Useful for Safety Performance Functions (SPFs): Poisson regression models are
used to develop SPFs, which predict crash frequencies based on factors like traffic
volume, road geometry, and environmental conditions. These models help agencies
estimate expected crashes and assess safety performance.
 Support for Empirical Bayes Methods: The Poisson model is integral to the
Empirical Bayes (EB) approach, which combines predicted and observed crash data to
better estimate expected crash frequencies. This method reduces the impact of random
fluctuations in crash data and supports evidence-based safety decisions.

The Poisson distribution plays a foundational role in traffic accident modeling due to
its suitability for rare, discrete events. It allows researchers and traffic engineers to estimate
crash probabilities, develop predictive models, and support proactive safety planning. While
simple, it serves as a gateway to more complex and realistic models in modern traffic safety
analysis.
The Poisson distribution is widely used in traffic crash data analysis to model the
frequency of crashes at specific locations over time. It is suitable for rare, discrete events and
serves as a foundational model in traffic safety studies. However, despite its usefulness, the
Poisson model has several limitations when applied to real-world crash data. Understanding
these limitations is crucial for selecting the appropriate analytical approach and improving the
accuracy of traffic safety evaluations.

Key Limitations of Using the Poisson Distribution:

1. Assumption of Equal Mean and Variance: The Poisson distribution assumes that the
mean and variance of crash frequency are equal. In practice, traffic crash data often exhibit
overdispersion, where the variance exceeds the mean. This is due to unobserved
heterogeneity — factors such as weather, driver behavior, or time-of-day effects that are not
accounted for in the model but still influence crash frequency.
Impact: Underestimating the variance can lead to understated standard errors, inflated
significance levels, and misleading inferences, potentially resulting in the identification of
non-existent safety issues or overlooking real ones.

2. Inability to Handle Excess Zeros: Crash data at many road segments or intersections
may show zero crashes during the observation period. The standard Poisson model does not
account for these excess zeros, leading to poor model fit and inaccurate predictions.
Impact: The model may overpredict crashes at locations with zero counts or underpredict at
high-crash locations, weakening its reliability in safety performance evaluations.

3. Assumption of Independent Events: The Poisson distribution assumes that crash events
are independent of each other. However, in real-world settings, crashes may be correlated due
to spatial (nearby locations) or temporal (within short periods) effects.
Impact: Ignoring such correlations can compromise the model's accuracy and reduce its

Department of Civil Engineering, ATMECE, Mysuru Page 6

Road Safety Engineering BCV755A

effectiveness in identifying high-risk locations or assessing safety interventions.

4. No Room for Unobserved Heterogeneity: The Poisson model assumes a homogeneous

process, meaning all locations or time periods are assumed to be governed by the same
underlying crash mechanism. In reality, different sites have unique characteristics affecting
crash risk, such as road design, traffic volume, and driver demographics.
Impact: Failure to incorporate heterogeneity may result in biased or inconsistent parameter
estimates.

Limitations Can Be Addressed:

Use of Negative Binomial (NB) Regression: To address overdispersion, analysts often
replace the Poisson model with a Negative Binomial (NB) model, which introduces a
dispersion parameter allowing the variance to exceed the mean. This model is more flexible
and better suited for real-world crash data.
Zero-Inflated and Hurdle Models: To handle excess zeros, Zero-Inflated Poisson (ZIP) and
Zero-Inflated Negative Binomial (ZINB) models are used. These models combine two
processes: one predicting the probability of zero crashes and another modeling positive crash
counts. Hurdle models are also applied when different processes govern zero and non-zero
crash occurrences.
Random Effects and Hierarchical Models: To account for spatial and temporal correlations
and unobserved heterogeneity, random-effects models or hierarchical Bayesian models are
used. These allow for varying effects across locations or time, improving accuracy and
generalizability.

While the Poisson distribution provides a simple starting point for crash data analysis,
it has significant limitations, especially when dealing with over dispersed data, excess zeros,
and unobserved variability. Advanced models like the Negative Binomial, zero-inflated
models, and random-effects frameworks offer more accurate and reliable tools for modern
traffic safety research and decision-making.

Problems
1. It is observed that on an average a vehicle driver drives 5000 km during the course of
a year. The probability of having an accident is 100 per 200 million vehicle-kilometres.
What is the probability of a driver having at least two accidents during his driving
career extending to 25 years?
Solution:
No. of accidents = 100 per 200,000,000 vehicle kilometres
p= =0.0000005 accidents per vehicle kilometre
∗
M = 5000 * 25 = 125000km
m = p M = 0.0000005 * 125000 = 0.0625

𝑚 𝑒
P(r) =
𝑟!

Department of Civil Engineering, ATMECE, Mysuru Page 7

Road Safety Engineering BCV755A

𝑃(𝑟 = 𝑎𝑡 𝑙𝑒𝑎𝑠𝑡 2) = 1 − 𝑃(𝑟 = 0) − 𝑃(𝑟 = 1)

. ∗ . . ∗ .
P(r) = 1 − −
! !

. .
P(r) = 1 − 𝑒 − 0.0625 ∗ 𝑒

P(r) = 0.012

2. The accident records for three consecutive years at an uncontrolled junction indicate
the following number of accidents:
Year No. of accidents
1972 3
1973 6
1974 9
Calculate the probability of 4 accidents occurring per year at the site.

Solution:
𝑚= =6
𝑚 𝑒
P(r) =
𝑟!
6 ∗𝑒 6 ∗𝑒
P(4) = =
4! 24
P(4) = 0.134

3. It has been found that on an average 1 in 100 drivers in a bus company are involved
in an accident every year. If there are 500 drivers in the company, what is the
probability that there are exactly 4 drivers who are involved in an accident during a
year?
Solution:
p= =0.01

m = n * p= 500 * 0.01 = 5
𝑚 𝑒
P(r) =
𝑟!

5 ∗𝑒 5 ∗𝑒
P(4) = =
4! 24
P(4) = 0.48

2.2.3 Chi- Squared Distribution

One of the situations a traffic engineer has to assess frequently is whether the safety
measures adopted at a particular location or stretch of road have been really effective in
reducing the number of accidents. Before and after data on accidents can be evaluated on
statistical principles, and one of the handy tools in this direction is the Chi-squared test.

Department of Civil Engineering, ATMECE, Mysuru Page 8

Road Safety Engineering BCV755A

Let b be the number of accidents before the improvements at a particular location and
a the number after the improvements. Assuming that the improvements have no effect and the
accident number is expected to increase due to the changes in traffic and weather then let b C
be the number of accidents expected if no improvements had been carried out, the factor C
being called the control ratio. Then the value of the Chi-squared is:
(𝒂 − 𝒃𝑪)𝟐
𝝌𝟐 =
(𝒂 + 𝒃)𝑪
The null hypothesis Ho stipulates that there is no real change due to the improvements.
Assuming a 5% level of significance, the value of χ2 to be 3.841 with one degree of
freedom. If χ2 > 3.841, we observe that the null hypothesis is unlikely to be true and that
there is a real change. On the other hand, if χ2 < 3.841 we conclude that the null hypothesis is
true and that there is no real change due to the improvements.

The chi-squared χ2 test is a very useful statistical tool and has many applications.
Following are the important applications to the traffic engineering. They are:
Testing of proportions with contingency tables: A convenient application of the χ2 test is
in testing the comparability of observed and expected values in two-way tables, known as
contingency tables

Goodness-of-fit test: Another useful application of the chi-squared distribution is in the

goodness-of-fit test. Under this test, the measure of the discrepancy between a set of observed
data and the values that are to be expected if the results follow a hypothesis distribution is
evaluated. For example, the analyst will have an answer to the following types of questions:

(i) It is observed from past experience that the spot speeds follow a normal distribution. Are
the data obtained from a particular study also in line with the previous experience?

(ii) The arrival pattern of vehicles is known to be generally random and follows the Poisson
distribution. Do the data obtained from a particular study also indicate a Poissonian arrival
pattern?

Benefits of Using the Chi-squared Test

 Simplicity and Clarity: The Chi-squared test is straightforward and easy to compute, making
it suitable for practical engineering evaluations.
 No Assumption of Normality: It is a non-parametric test and doesn’t require data to follow a
normal distribution.
 Objective Evaluation: It provides an evidence-based way to assess whether safety
improvements are effective, supporting data-driven decision-making.

Limitations and Considerations:

 The test requires a sufficient sample size; small numbers can produce unreliable results.
 It only tests association, not causation.
 It assumes that the expected frequencies are based on a valid and consistent baseline.
The Chi-squared test serves as a valuable analytical tool in traffic engineering for evaluating
the statistical significance of changes in crash rates following the implementation of safety

Department of Civil Engineering, ATMECE, Mysuru Page 9

Road Safety Engineering BCV755A

countermeasures. By comparing observed and expected crash frequencies, it helps determine whether
safety interventions have had a meaningful impact, thereby supporting the continuous improvement of
roadway safety strategies.

Problems
1. In an ordinary square junction of two roads there were 20 accidents in a year. After
provision of traffic signals, the number of accidents dropped down to 8 per year. In the
sector of the city where this junction is situated, the general trend observed was that
number of accidents increased at a rate of 10 per cent during the period covered by the
above two observations. Test whether the improvement in junction design has a
significant effect at 5% significance level.
Solution:
Here, a = 8; b = 20; C = 10 % increase

110
𝐶= = 1.1
100
( ) ( ∗ . )
𝜒2 = ( )
= ( )∗ .

𝜒2 = 6.36

χ2 for 5% significance level and one degree of freedom = 3.841

As χ2 (observed) > 3.841, the data supplied provides strong evidence that the improvement in
junction design was effective, and that the reduction in the number of accidents was not
merely due to chance alone.

2. The accident data pertaining to a metropolitan city for the year 1965 and 1970 are
given below:
Year 1965 1970
Accidents 300 400
Vehicle-kilometre of travel 250 million 300 million
Test whether there is any significant increase in the accident rates in two
Solution:
Here, the traffic has increased from 250 to 300 million vehicle kilometres, and thus the
control rate C
300
𝐶= = 1.2
250

Here, a = 400; b = 300

( ) ( ∗ . )
𝜒2 = ( )
= ( )∗ .

𝜒2 = 1.90

Department of Civil Engineering, ATMECE, Mysuru Page 10

Road Safety Engineering BCV755A

χ2 for 5% significance level and one degree of freedom = 3.841

As χ2 (observed) < 3.841

Conclude that the difference in the accident rates in the years might easily have arisen
due to chance.

2.2.4 Statistical Comparison

In road safety analysis, statistical comparison is a valuable tool used to identify and analyze
differences in crash data.
Here are some short notes on statistical comparison in road safety analysis:
Types of statistical tests: There are various statistical tests that can be used to compare crash
data, such as t-tests, chi-square tests, ANOVA, and regression analysis. The choice of test
depends on the type of data and research question being investigated.
Data requirements: Statistical tests require large enough sample sizes to ensure that the
results are reliable and accurate. Additionally, the data must be representative of the
population being studied and must meet certain assumptions for the chosen statistical test.

Comparing crash data: Statistical comparison can be used to analyze differences in crash
data between different geographic regions, time periods, or demographic groups. This
information can help identify high-risk areas and populations that require targeted safety
interventions.

Limitations: It is important to recognize the limitations of statistical comparison in road

safety analysis.

For example, statistical tests can only identify correlations, not causation. Additionally, the
results may be influenced by other factors not accounted for in the analysis.
Overall, statistical comparison is a powerful tool for analyzing crash data in road
safety analysis.
However, it is important to use appropriate statistical tests and to interpret the results
carefully, taking into account the limitations of the analysis.

Types of Statistics used in Traffic Studies

1. Descriptive statistics: This method involves the use of summary statistics such as mean,
median, mode, standard deviation, and percentiles to describe the characteristics of traffic
safety data. This method helps to identify patterns and trends in the data.
2. Inferential statistics: This method involves making inferences about the population based
on a sample of data. This method includes hypothesis testing, confidence intervals, and
regression analysis.
3. Time series analysis: This method involves the analysis of data collected over time to
identify trends and patterns. Time series analysis is used to predict future trends and can help
in identifying seasonal patterns in traffic safety data.

Department of Civil Engineering, ATMECE, Mysuru Page 11

Road Safety Engineering BCV755A

4. Cluster analysis: This method involves grouping similar data points into clusters. This
method is useful for identifying clusters of high accident areas, which can help in developing
targeted interventions to improve traffic safety.
5. Spatial analysis: This method involves the analysis of traffic safety data in a spatial
context. This method includes mapping and spatial regression analysis. Mapping can help in
identifying spatial patterns of traffic accidents, and spatial regression analysis can help in
identifying the factors that contribute to accidents in specific areas.
6. Multivariate analysis: This method involves the analysis of multiple variables
simultaneously. This method is useful for identifying the factors that contribute to traffic
accidents and for predicting the likelihood of future accidents based on these factors.
Multivariate analysis includes techniques such as factor analysis and principal component
analysis.

2.3 Traffic Management Measures and Their Influence on Accident Prevention.

2.4 Recommended questions

1. Explain how regression analysis is used in traffic safety studies. What are the benefits
of using regression models for predicting accident frequency at specific locations?
2. Discuss the application of the Poisson distribution in modeling accident occurrences.
Why is the Poisson distribution commonly used in traffic accident studies?
3. What are the limitations of using a Poisson distribution in traffic crash data analysis,
and how can these be addressed?

2.5 Outcomes
 Evaluate the effectiveness of road safety interventions and conduct post-
implementation analysis.

2.6 Further Reading

 [Link]
 [Link]
ml
 [Link]
 [Link]
[Link]

Department of Civil Engineering, ATMECE, Mysuru Page 12