Application of Negative Binomial Distribution to Accident Analysis and

Prevention

The negative binomial distribution has become a topic of great interest to

American actuaries in the last few years. A lot of researches have been done to
discover what negative binomial is, where does it come from and what does it mean.
A negative binomial distribution (also called the Pascal Distribution) is a discrete
probability distribution for random variables in a negative binomial experiment. The
random variable is the number of repeated trials, X, that produce a certain number of
successes, r. In other words, it is the number of failures before a success. This is the
main difference from the binomial distribution: with a regular binomial distribution,
we usually look at the number of successes. With a negative binomial distribution, it
is the number of failures that counts.

Some questions may still remain or some new ones brought to mind, but one
thing for sure, negative binomial distribution has many real-life applications
especially in the accident analysis and prevention terminology. According to the
terminology, to derive this distribution, we should make an assumption that everyone
starts out with the same propensity toward having an accident which remains constant
until an accident occurs. When an accident does occur, the future probability for that
individual is changed. This development also leads to a negative binomial curve. This
terminology leads to a very good discussion of the bivariate negative binomial and its
relationship to accident proneness.

Greenwood and Yule1 presented the basic paper which developed the theory into
a mathematical model and tested it on actual accident data. This paper is a classic and
is referred to by many authors. For example, Kendall and Stuart2 summarize this
paper very well and give the data that Greenwood and Yule used. Kendall and Stuart
develop the negative binomial in two ways. The more interesting method is in

1
Greenwood, M. and Yule, G. Udny, "An inquiry into the nature of frequency distributions representative of multiple
happenings with particular reference to the occurrence of multiple attacks of disease or of repeated accidents," Journal of

the Royal Statistical Society 83:255-279, 1920

2
Kendall, M. G. and Stuart, A., The advanced theory of statistics, Vol. 1. New York: Hafner Publishing Co., pp. 129, 225,
1958

1
discussing sequential sampling when the objective is to continue sampling until a
certain number of successes has been achieved. The number of items sampled will
then follow a negative binomial distribution. During the ensuing years, a number of
authors followed this 1920 development and utilized this curve in describing accident
phenomena. They dealt primarily with industrial accidents and were concerned, in
many cases, with the psychological and sociological problems connected with the
accident proneness phenomenon. If different people had a different accident
proneness, we might improve safety if we could detect it or if we could change it.

There are numerous applications of the negative binomial distribution on accident

analysis and prevention in the literature. Negative binomial is mainly used to predict
and analyse accident proneness. This is important in many countries as death rate
increases mainly due to crash. This explains why there are several papers discussing
this terminology and different issues were brought up.

One of the papers that were chosen is entitled ‘A mixed effects negative binomial
analysis of road mortality determinants in Sub-Saharan African countries’ 3 .
According to this journal, the study is referring to the road safety in African regions
which has the least safe road in the world. It is proven by the analytical approach of
negative binomial distribution because of the prevalent over-dispersion in crash data.
This study investigates the potentials predictors of road mortality for a panel of 23
Sub-Saharan African countries. This paper uses aggregrate data for a panel of 23
Sub-Saharan African countries to explore the potential predictors of road traffic
fatalities between 2001 and 2010. It states that road traffic accidents are the cause of
around 1.2 millions deaths and 50 million injured worldwide every year. These
numbers are expected to rise by almost 65% in the coming decades if no preventive
action is undertaken. This study is therefore conducted as an initial attempt to explore
the factors associated with traffic crashes and death toll in USA. This assumptions
claims that traffic fatalities increase with the national income in non-develop
countries. The selection of the variables included in the analysis was based on an
extensive review of previous aggregate studies which identified many determinants of
traffic fatalities. A primary inspection of the data set revealed no risk of

3 Wonmongo Lacina Soro, Didier Wayoro, A mixed effects negative binomial analysis of road mortality
determinants in Sub-Saharan African countries, Transportation Research Part F: Traffic Psychology and Behaviour,
Volume 52, 2018, Pages 120-126.

2
multicollinearity. The models simultaneously incorporate the fixed and the random
effect. The road mortality in the study is defined by the dependent variable. The data
set covers from 2001 until 2010. In this situation the researchers mixed the negative
effect of negative binomial distribution. It involves model 1, model 2 and 3. All of the
three model consist of different type of negative binomial approaches whereby model
1 is a random effect of negative binomial, model 2 is a fixed effects negative binomial
and model 3 is an estimation of robust standard errors. The non linear relationship
between the GDP per capita and traffic fatalities is tested by using regression
techniques. This is because it has a negative coefficient while its square has a positive
one. Through this study, it is found that traffic fatalities increase with the national
income level proxied by the Gross Domestic Product per capita as mentioned in the
assumption earlier. Therefore we can agree that the negative binomial distribution in
this situation gives a good fit of either distributions and high frequency of occurence
as the coefficient for each parameter are consistent across all models.

Another application of negative binomial distribution in the accident analysis and

prevention terminology is the application of a random parameters Negative Binomial
Lindley model to examine multi-vehicle crashes along rural mountainous highways in
Malaysia. The objective of this study was to examine multi-vehicle crashes along
rural mountainous highways by developing an appropriate safety performance
function. Challenges for this research are the availability of reliable and accurate data,
and the develop an appropriate Safety Performance Functions (SPF) by accounting for
heterogeneity resulting from excess zero counts. To overcome these challenges, firstly
an extensive data collection task was undertaken to assemble a comprehensive set of
exogenous factors. Next, the data is tested and compared with the performance of a
Random Parameters Negative Binomial (RPNB) model and the state of-the-art
Random Parameters Negative Binomial-Lindley (RPNB-L) models with excess zeros.
Somehow, this research has identified that the RPNB-L model is a better option than
RPNB and Random Parameters Negative Binomial-Generalized Exponential
(RPNB-GE) models in dealing with excess zeros. This result is very promising as the
RPNB-L model can deal with various other issues related to heterogeneity. To
reconfirm this finding, the RPNB-L model should be tested using other datasets with
excess zeros. The parsimonious RPNB-L model identified six significant variables
affecting multi-vehicle crashes along rural mountainous roads in Malaysia. While

3
exposure (ADT x segment length), heavy rainfall at the time of crash and number of
minor junctions were positively associated with multi-vehicle crashes, the presence of
horizontal curves along a steep gradient, the presence of a passing lane and the
presence of road delineation were negatively associated. Among these, presence of
horizontal curves along a steep gradient and the presence of road delineation were
identified as random parameters, suggesting unobserved heterogeneity around these
two parameters in explaining multi-vehicle crashes. Although the dataset used in this
study is from 2008 to 2012, the findings of this study provide considerable insights
into the factors affecting multi-vehicle crashes along rural mountainous highways in
Malaysia, which would be helpful in developing effective countermeasures to
improve road safety in rural mountainous areas in developing countries.

In another journal written by students of Monash University and University of

Melbourne, negative binomial distribution is used in examining tram-involved crash
frequency on route sections in Melbourne, Australia. This research was used to
determine how many crash of vehicles happen and how to prevent them from
happening again. Public transport is becoming more important as mobility,
accessibility, and environmental problems are increasing in cities around the world
(Vuchic, 1981, Gakenheimer, 1999; Fouracre et al., 2003)4. However trams present a
range of inherent safety issues regarding their design and operational characteristics,
since they are large and heavy vehicles operating in confined, mixed and complex
environments with pedestrians and cars. Cheung et al. (2008)5 developed zonal
level transit-involved collision prediction models for urban transit in Toronto, Canada
using a negative binomial regression structure so that the relative contribution of
factors influencing tram-involved collisions are clear. Both tram and bus-involved
collisions are combined to identify transit-involved crash causation factors. The
results show that vehicle kilometers travelled, transit kilometers travelled, arterial
road length, transit stop density, percent of near side stops and average posted speed
are significant indicators for transit-involved collisions. Shahla et al. (2009) 6
developed a negative binomial crash prediction model for intersections in Toronto,

4 Vuchic, V.R., 1981. Urban Public Transportation; Systems and Technology. Prentice-Hall
5 Cheung, C., Shalaby, A.S., Persaud, B.N., Hadayeghi, A., 2008. Models for safety analysis of road surface
transit. Transp. Res. Rec.: J. Transp. Res. Board 2063,168–175.
6 Shahla, F., Shalaby, A.S., Persaud, B.N., Hadayeghi, A., 2009. Analysis of transit safety at signalized
intersections in Toronto, Ontario, Canada. Transp. Res. Rec.:J. Transp. Res. Board 2102, 108–114.

4
Canada using five years of transit collision data and pointed out that annual average
daily traffic (AADT), public transit and pedestrian traffic volumes, turn movement
treatments, public transit stop location, mode technology and availability of transit
signal priority technology have significant associations with public transit related
collisions at signalized intersections.

So, the first step in using negative binomial in this research was to is to identify
the key traffic, transit and route factors that influence tram-involved crash frequencies
along tram route sections using these formula to find it.

E(A) represents the predicted number of crashes per year

where E(Ait) represents the predicted number of crashes along tram route section
i in year t.

The results showed that tram service frequency per week, tram stop spacing, tram
route section length, tram signal priority, general traffic volume and tram lane priority
have considerable association with tram-involved crash occurrences. The average
speed of the tram and the proportion of platform tram stops are also linked to
tram-involved crash frequency but at a lower level in effect size.

The study findings provide useful insights on route section level tram-involved
crash occurrence and present useful planning tools for transit agencies. There are
numerous opportunities to carry out further research in this field. In particular, further
col- lection of tram incident data could be undertaken as a means to improve model
validity and possibly identify other factors that can be significant in explaining
tram-involved crash frequency. Also more advanced modelling approaches can be
adopted as a way to improve model fitness. Identifying key factors associated with
different crash severity levels will also be a potential area for further research.

5
 By using the negative binomial distribution, accident proneness can easily be
predicted in certain countries. This distribution is applied in a study which predicts
expressway crash in China. The expressway crash frequency in China is predicted
using a random effect negative binomial. This is being discussed in a journal written
by Zhuanglin Ma, Honglu Zhang, Steven I-Jy Chien, Jin Wang and Chun JiaoDong7.
The main objective of this journal is to investigate the relationship between crash
frequency and potential influence factors for the crash to occur in China. Another
motive of this research is to access the performance of negative binomial and random
effect negative binomial model since negative binomial model and a random effect
negative binomial were developed to predict crash frequency. The researchers are also
determined to investigate which kind of road segment method is suitable to the
straight and flat area expressway. The parameters of both models were determined
using the maximum likelihood(ML) method, and the mixed stepwise procedure was
applied to examine the significance of explanatory variables. It is found that
longitudinal grade, road width, ratio of longitudinal grade and curve radius were
founded as significantly affecting crash frequency. The model performance was
determined by the relative prediction error and the cumulative standardized residual.

The comparison between predicted crash frequencies of the negative binomial

and random effect negative binomial models on the basis of fixed length segmentation
is studied. It was found that the predicted crash frequency from the RENB model is
closer to the actual crash frequency. The relative prediction error of the negative
binomial model is larger than of the random effect negative binomial model. By the
same token, one can find that the cumulative standardized residual of the negative
binomial models is greater than random effect negative binomial model, also indicate
that the random effect negative binomial model outperforms the negative binomial
model in terms of prediction accuracy. The performance of the negative binomial and
the random effect negative binomial models are also compared based on homogenous
longitudinal grade segmentation and the crash frequency was predicted using the ML
method. A mixed stepwise procedure was used to examine the significance of
explanatory variables under the hypothesis of a 95% confidence interval, and the
7
Zhuanglin Ma, Honglu Zhang, Steven I-Jy Chien, Jin Wang, Chunjiao Dong, Predicting expressway crash
frequency using a random effect negative binomial model: A case study in China, Accident Analysis & Prevention,
Volume 98, 2017, Pages 214-222, ISSN 0001-4575, https://doi.org/10.1016/j.aap.2016.10.012.

6
goodness-of-fit as the prediction measures is calculated. The random effect negative
binomial model is superior to the negative binomial model, based on not only on the
fixed-length segment, but also on the homogenous longitudinal grade method.
Longitudinal grade and road width have significance effects on crash frequency for
the random effect negative binomial model based on both the fixed-length and
homogenous longitudinal grade method.

Another similar case study is done in China which is to examine the crash
frequency for the freeways in China. This study was expected to provide a better
understanding of how traffic condition and freeway design affect safety and should be
useful to freeway engineers to design safe freeways, or develop effective safety
countermeasures. However, there are unacceptable safety issues and high crash rate,
which draw more and more attentions and concerns from both the road authorities and
drivers. In the aspect of analytical methodology for crash frequency, negative
binomial (NB) model was widely used as they can properly represent crash counts.
The individual effects can be defined as fixed or random resulting in random effects
negative binomial (RENB) model. This research is to better understand the safety
effects of traffic condition and freeway design elements using a RENB model together
with a traditional NB model. A wide range of crash frequency models have been
developed over the past decades to gain the knowledge of factors contributing to crash.
Due to the nonnegative integer nature of crash and the over-dispersion characteristics
of crash data, negative binomial (NB) model was extensively used in the past studies.
The data of crash, traffic volume and freeway design were collected and assembled
for this study. Crashes of total number of 22,791 were obtained along the 1542 km of
the studied eight freeways from Guangdong Provincial Freeway Administration
(GDFA) and Liaoning Provincial Freeway Administration (LNFA). The annual traffic
volume data covering the data periods were counted by volume observation stations
on the corresponding freeways and provided by GDFA and LNFA. A backward
stepwise regression technique was used to obtain the parsimonious model with the
best subset of estimated parameters. The insignificant variables were removed one by
one and the significant ones at 95 confidence level were kept in the model. Note that
Marginal effects were the average effects of a unit increase in an independent variable
on the annual crash frequency on an average segment in this study. The estimated NB
and RENB models all included 10 significant variables and most of estimated

7
parameters are of reasonable magnitudes with plausible signs. The dispersion
parameter α was significantly different from zero, indicating that the variance of crash
data was larger than the mean and that the NB model was a justified choice. The
proportion of trucks was positively associated with crash frequency. This result
indicated that an increase in truck percentage was associated with increased crashes,
which was consistent with our expectations and previous studies 89. Trucks tend to
move at a relative lower speed than other vehicles resulting in an increasing
propensity for lane-changing and passing maneuvers 1011. Moreover, heavy trucks
increase the risk for rear-end crashes due their high inertia and reduced braking
performance. The other achievements of this research is the contribution of model
evaluation and that many factors significantly affecting safety were identified and
their effects on crash frequency were quantified by their marginal effects. Two
exposure variables were positively related with crashes. The crash probability
increased with the increase in truck percentage due to the increased speed difference
between trucks and cars. Therefore the results from this study may shed lights on
effects of traffic and freeway design on safety. The findings are useful for the safety
design and management of freeways and expected to provide references for
developing safety countermeasures for freeways, in particular for freeways in China.

8
L. Garach, J. de Oña, G. López, L. Baena, Development of safety performance functions for spanish two-lane
rural highways on flat terrain, Accid. Anal.

Prev. 95 (2016) 250–265.

9
M. Hosseinpour, A.S. Yahaya, A.F. Sadullah, Exploring the effects of roadway characteristics on the frequency
and severity of head-on crashes: Case studies from Malaysian Federal Roads, Accid. Anal. Prev. 62 (2014) 209–
222.
10T.Q. Tang, J. Zhang, L. Chen, H.Y. Shang, Analysis of vehicle’s safety envelope under car-following model,
Physica A 474 (2017) 127–133.
11W.X. Zhu, L.D. Zhang, A speed feedback control strategy for car-following model, Physica A 413 (2014) 343–
351.

8
BIBLIOGRAPHY

5 main journals used:

1. Wonmongo Lacina Soro, Didier Wayoro (2018). A mixed effects negative

binomial analysis of road mortality determinants in Sub-Saharan African countries,
Transportation Research Part F: Traffic Psychology and Behaviour, 52, (120-126).
https://doi.org/10.1016/j.trf.2017.11.024

2. Rusdi Rusli, Md. Mazharul Haque, Amir Pooyan Afghari & Mark King (2018).
Applying a random parameters Negative Binomial Lindley model to examine
multi-vehicle crashes along rural mountainous highways in Malaysia. Accident
Analysis & Prevention, 119, 80 - 90. https://doi.org/10.1016/j.aap.2018.07.006

3. Farhana Naznin, Graham Currie, David Logan & Majid Sarvi (2016). Application
of a random effects negative binomial model to examine tram-involved crash

9
frequency on route sections in Melbourne, Australia. Accident Analysis & Prevention,
92, 15 - 21. https://doi.org/10.1016/j.aap.2016.03.012

4. Zhuanglin Ma, Honglu Zhang, Steven I-Jy Chien, Jin Wang, Chunjiao Dong (2017).
Predicting expressway crash frequency using a random effect negative binomial
model: A case study in China. Accident Analysis & Prevention, 98, 214-222,
Ihttps://doi.org/10.1016/j.aap.2016.10.012.

5. Qinzhong Hou, Xianghai Meng, Junqiang Leng & Lu Yu (2018). Application of a

random effects negative binomial model to examine crash frequency for freeways in
China. Physica A: Statistical Mechanics and its Applications, 509, 937 - 944.
https://doi.org/10.1016/j.physa.2018.06.088

Other references:

1. Greenwood, M. and Yule, G. Udny (1920). "An inquiry into the nature of
frequency distributions representative of multiple happenings with particular
reference to the occurrence of multiple attacks of disease or of repeated accidents,"
Journal of the Royal Statistical Society 83:255-279

2. Kendall, M. G. and Stuart, (1958). The advanced theory of statistics, Vol. 1. New
York: Hafner Publishing Co., pp. 129, 225

3. Vuchic, V.R., (1981). Urban Public Transportation; Systems and Technology.

Prentice-Hall.

4. Cheung, C., Shalaby, A.S., Persaud, B.N., Hadayeghi, A., (2008). Models for
safety analysis of road surface transit. Transp. Res. Rec.: J. Transp. Res. Board 2063,
168–175.

5. Shahla, F., Shalaby, A.S., Persaud, B.N., Hadayeghi, A., (2009). Analysis of transit
safety at signalized intersections in Toronto, Ontario, Canada. Transp. Res. Rec.:J.
Transp. Res. Board 2102, 108–114.
6. L. Garach, J. de Oña, G. López, L. Baena, Development of safety performance
functions for spanish two-lane rural highways on flat terrain, Accid. Anal.
Prev. 95 (2016) 250–265.

10
7. M. Hosseinpour, A.S. Yahaya, A.F. Sadullah (2014). Exploring the effects of
roadway characteristics on the frequency and severity of head-on crashes: Case
studies from Malaysian Federal Roads, Accid. Anal. Prev. 62, 209–222.

8. T.Q. Tang, J. Zhang, L. Chen, H.Y. Shang (2017). Analysis of vehicle’s safety
envelope under car-following model, Physica A, 474, 127–133.

9. W.X. Zhu, L.D. Zhang (2014). A speed feedback control strategy for car-following
model, Physica A 413, 343–351.

11