Use of Poisson Distribution

in Highway Traffic

Daniel L. Gerlough

The Probability Theory Applied

to Distribution of Vehicles

on Two-Lane Highways

Andre' Schuhl

THE ENO FOUNDATION FOR HIGHWAY TRAFFIC CONTROL

SAUGATUCK . 1955 - CONNECTICUT

Copyright 1955, by the Eno Foundation for Highway Traffic Control, Inc. All
rights reservedunder Internationaland Pan-American CopyrightConventions

Library of Congress catalog card: 55 -11902

Printed in the United States of America by Columbia University Press

 Foreword

These two mathematical discussions of traffic analyses tend to

confirm the theory that if we are properly to understand traffic
we must learn more about its behavior; and that if we expect
to control traffic efficiently and provide suitable means for con­
trol, we must be able to uniformly predict traffic patterns.
To know what to expect will provide a guide to a rule of
action to fit the problem.
The authors, Mr. Gerlough and Mr. Schuhl, live in distantly
separated parts of the world. It is safe to assume they were
motivated in preparing their articles by their daily interest and
activities in traffic.
They are highly competent and have that very valuable ad­
vantage of experience so essential in translating theory to
practice.
Continued efforts in this direction tend to provide an ac­
cumulation of basic knowledge of traffic performance and in
time to evolve new concepts of our traffic problem with im­
proved ideas in planning and control.
Trusting that this brochure will be some encouragement to
further research, the Eno Foundationwelcomes the opportunity
to publish it and wishes to express its appreciation to the
authors.
ENo FoUNDATION
 Use of Poisson Distribution
in Highway Traffic

DANIEL L. GERLOUGH

Mr. Gerlough, a Registered Professional Electrical

Engineer in the state of California, is an engineer
with the Institute of Transportation and Traffic
Engineering at the University of California at Los
Angeles. For ten years prior to his joining the In­
stitute at Los Angeles in 1948, he served as an engineer
for six corporations in Texas and California. Mr.
Gerlough is a member of the Institute of Radio
Engineers, of the American Institute of Electrical
Engineers, and of the Institute of Traffic Engineers.

Applications of probability and statistics to engineering prob­

lems have increased markedly during the past decade. These
techniques are particularly helpful in the study of vehicular
traffic. A monograph' published recently by the Eno Founda­
tion presents applications of the broad field of statistics to
traffic problems. The present paper discusses in detail one par­
ticular probability distribution which has been found useful in
the treatment of several types of problems encountered by the
traffic engineer.
The Poisson distribution is a mathematical relationship
which finds such applications as:
Testing the randomness of a given set of data
Fitting of empirical data to a theoretical curve
Prediction of certain phenomena from basic data
Greenshields, Bruce D., and Weida, Frank M., Statistics with Applications
to Highway Traffic Analyses, The Eno Foundation, 1952.
1
2 POISSON DISTRIBUTION

Some of the uses open to the traffic engineer include:

Analysis of arrival rates at a given point

Studies of vehicle spacing (gaps)
Determination of the probability of finding a vacant park­
ing space
Studies of certain accidents
Design of left-turn pockets

The application of the Poisson distribution to traffic prob­

lems is not new. Certain applications were discussed by Kinzer2
in 1933, Adams3 in 1936, and GreenshieldS4 in 1947- It is the
purpose here to provide a more extensive treatment in a form
that will serve as a reference guide for both students and
practitioners of traffic engineering. In attempting to fulfill this
objective the following material has been included:

Examples which cover typical cases and at the same time indicate
the variety of applicability.
Detailed descriptions of the steps to be followed in applying the
Poisson distribution and in testing it by theX2 (chi-square) test.
Charts for simplified computations as well as references to stand­
ard tables and other computational aids.
Derivation of the Poisson distribution (in sections B to E of the
Appendix) starting from the fundamental concepts of permutations
a n d c o m b i.- --,. iwns.
Bibliographical references to more advanced applications in
which the Poisson distributionis combined with other relationships.

2 Kinzer, John P., Application of the Theory of Probability to Problems of

Highway Traffic, thesis submitted in partial satisfaction of requirements for
degree of B.C.E., Polytechnic Institute of Brooklyn, June 1, 1933.
Abstracted in: Proceedings, Institute of Traffic Engineers, vOl. 5, 1934, Pp. i i8­
124.
3 Adams, William F., "Road Traffic Considered as a Random Series," journal
Institution of Civil Engineers, VOL 4, Nov.,'1936, pp. 121-130+.
4 Greenshields, Bruce D.; Shapiro, Donald; Ericksen, Elroy L., Traffic Per­
formance at Urban Street Intersections, Technical Report No. i, Yale Bureau
of Highway Traffic, 1947.
 POISSON AND TRAFFIC 3

Nature of Poisson Distribution

When events of a given group occur in discrete degrees (heads

or tails, 1 to 6 on the face of a die, etc.) the probability of occur­
rence of a particular event in a specified number of trials may
be described by the Bernoulli or binomial distribution (See
Appendix, Sections B and Q.
As an example, let us consider an experiment consisting
of five successive drawings of a ball from an urn contain­
ing uniformly mixed black and white balls, with the drawn
ball being returned to the urn after each drawing. Of the five
drawings which comprise a single experiment, let x be the
number which produced black balls; thus x can equal o, 1, 2, 3,
4, or 5. Let P(x) be the probability that in a given experiment
the number of black balls drawn would be exactly x. If p is the
probability that a particular drawing will yield a black ball,
and q(q i-p) is the probability that a particular drawing
yields a white ball, then

P(x) = Qp q-:'

where C.' is the number of combinations of 5 things taken x at

a time. It can be seen that in the above example p, the prob­
ability that a single drawing yields a black ball, is equal to
the percentage of black balls in the urn. With the ball being
replaced after each drawing p will remain constant from draw­
ing to drawing.
The experiment just described is an example of the
"Bernoulli" or "binomial" distribution in which P(x), the prob­
ability of exactly x successes out of n trials of an event where
the probability of success remains constant from event to event,
is given by

P(x) = CnPxqn-0

if the number of items in the sample n, becomes very large

while the product Pn
m is a finite constant, the binomial
4 POISSON DISTRIBUTION

distribution approaches the Poisson distribution as a limit.*

lim MXe__M
n--+ co P(x) = X!
Pn = m
where e is the Napierian base of logarithms.
The derivation of the Poisson distribution as a limiting case
of the binomial distribution is given in Appendix D. (The
Poisson distribution can be derived independently of the bi­
nomial distribution by more advanced concepts. See Fry.5)
The mathematical conditions of an infinite number of trials
and an infinitesimal probability are never achieved in practical
problems. Nevertheless, the Poisson distribution is useful as
approximating the binomial under appropriate conditions.
For such practical purposes, then, the Poisson distribution
may be stated as follows:
If in a given experiment the number of opportunities for an
event to occur is large (e.g., n '
- 50)
and
If the number of times a particular event occurs is small
(e.g., P -e- o.i)
and
If the average number of times the event occurs has a finite
value, m, (m
np),

Then Xx) = mze

xi
where x o, 1, 2....
In this statement of the Poisson distribution an experiment
may consist of such things as:
a. Observing the number of micro-organisms in a standard sample
of blood, x representing the number of micro-organisms in any
one sample, and n representing the number of samples studied.
b. Observing the number of alpha particles emitted during each
This implies that the probability of occurrence, p, becomes very small.
5 Fry, Thornton C., Probability and Its Engineering Uses, Van Nostrand, i928,
PP. 220-227.
 POISSON AND TRAFFIC 5
successive interval of t seconds. The number of such intervals
will be n, andX1, X2 ... x,, will be the number of particles dur­
ing the ISt, 2nd.... nth intervals.
c. Observing the number of blowholes in each of n castings, x
representing the number of holes in any one casting.

Examples of Poisson Distributions

The first record of the use of the Poisson distribution to treat
populations having the properties listed is attributed to
Bortkewitsch who studied the frequency of death due to the
"kick of a horse" among the members of ten Prussian Cavalry
Corps during a period Of 20 years. Some of the earliest engineer­
ing problems treated by the Poisson distribution were telephone
problems. The following example is based on such data:6
Example I
CONNECTIONS TO WRONGNUMBER

Number of wrong Number of periods exhibiting

connections per the given number of wrong
period connections (observed frequency)
0 0
1 0
2 1
3 5
4 11
5 14
6 22

7 43
8 31
9 40
10 35
11 20
12 i8
13 12
14 7
15 6
i6 2

6Thomdike, Frances, "Applications of Poisson's Probability Summation,"

Bell System Technical journal, Vol. 5, no. 4, Oct., 1926, pp. 604-624.
6 POISSON DISTRIBUTION

The fitting of the Poisson distribution to the experimental data

may be carried out in tabular form as follows:

Col. I Col. 2 Col- 3 Col- 4 Col- 5

X

Number of
wrong Observed Total
connections frequency wrong PM Theoretical
per period (periods) connections Probability frequency
0 0 0 0.000 0.0
I 0 0 0.001 0.3
2 1 2 o.oo6 i.6
3 5 15 o.oi8 4.8
4 11 44 0-039 10.4
5 14 70 o.o68 i8.2
6 22 132 0.099 26.4
7 43 501 0.124 33.1
8 31 248 0-135 36.o
9 40 36o 0-132 35.2
10 35 350 0.115 30-7
11 20 220 0.091 24-3
12 i8 2i6 o.o67 17.9
13 12 j56 0.045 12.0
14 7 98 0.028 7-5
15 6 go o.oi6 4.3
i6 2 32 0.009 2-4
> i6 0 0 0.007 1.9
TOTAL 267 2334 1.000 267.0
The entries in Col. 3 are the products of the corresponding entries
in Col. i and Col. 2.
In average number of wrong connections per period
2334
8-742
267

PW = (8-742)xe-8.742
X!

In the last row of the table (for more than i6 wrong connec­
tions per period) the values of probability and theoretical fre­
quency are necessary to make the totals balance. These values
represent the summation from 17 to infinity.
 POISSON AND TRAFFIC 7
Methods of evaluating e-0 x!, and P(x) are discussed later in
the paper.
Calculated Frequency
(Total periods observed) P(x)

2 67 P(x)
It will be seen that there is a high degree of agreement be­
tween the observed and calculated frequencies.
Note: Column 5 shows the theoretical or calculated fre­
quencies. When these frequencies are calculated, fractional
(decimally values often result. Here these values have been
rounded-off to the nearest o.i. The observed frequencies will,
of course, always be integral numbers.
Following the pioneer work in the field of telephone appli­
cations, the Poisson distributionwas gradually applied to other
engineering problems. The following example adapted from
Grant7 shows an application to the occurrence of excessive
rainfall:

Example 2

RAINSTORMS

Number
of storms
per 0 bserved Theoretical
station number of Total number of
peryear occurrences storms Probability occurrences
0 102 0 0-301 99-3
1 114 114 0-36i 119.1
2 74 148 0.217 7IA
3 28 84 o.o87 28.7
4 10 40 0.026 8.6
5 2 10 o.oo6 2.0
> 5 0 0 0.002 0.7
TOTAL 330 T96- 1.000 330.0

M
396
1.20 e-m
0.301 33o e-In
99.3
330
The first published examples of the Poisson distribution as
7Grant, Eugene L., "Rainfall Intensities and Frequencies," Transactions,
American Society of Civil Engineers, vol. 103, 1938, PP. 384-388.
8 POISSON DISTRIBUTION

applied to traffic data are presented by Adams.8 The following

is one of Adams' examples:
Example 3

RATE OF ARRIVAL (Vere St.)

(Number of vehicles arriving per io second interval)

No.
vehicles
per -ro
sec.
period Observed Total Probability Theoretical
X frequency vehicles P(X) frequency
0 94 0 0-539 97.0
1 63 63 0-333 59-9
2 21 42 0-103 i8.5
3 2 6 0.021 3.8
> 3 0 0 0.004 o.8
TOTAL i8o III 1.000 i8o.o
tn total vehicles III .617; e-.61-7
iO-talper
iods
_8o 0 539
P(X) Ltz)f e-G17
(.539) (.617)z
X! X!
Calculated frequency
18o P(x)
Note: Since there were I I I vehicles in I8o ten-second periods,
the hourly volume was 222.

Test-mg Goodness ot Fit (XI 'I'est)

In each of the foregoing examples it has been postulated that a
Poisson distribution having a parameter m whose value has
been computed from the observed data describes the popula­
tion that has been sampled. The observed distribution con­
stitutes this sample. By inspection there is apparent agreement
between the observed distribution (sample) and the theoretical
distribution. The inference is then made that the postulated
theoretical (Poisson) distribution is in fact the true population
8Adams, op. cit.
 POISSON AND TRAFFIC 9
distribution. This inference is based, however, solely on inspec­
tion; a more rigorous basis for reaching such a conclusion is
desired. One of several statistical tests of significance may be
used for this purpose; the chi square (X2) test is appropriate to
the present application. This test, which is described in Ap­
pendix G and illustrated below, provides for one of two
decisions:
i. It is not very likely that the true distribution (of which the ob­
served data constitute a sample) is in fact identical with the
postulated distribution.
2. The true distribution (of which the observed data constitute a
sample) could be identical with the postulated distribution.
It can be seen that either decision can be erroneously made.
Decision i can be wrong if in fact the postulated distribution
is the true distribution. On the other hand, Decision 2 can be
wrong if the true distribution is in fact different from the pos­
tulated distribution. Statistical tests of significance allow for
specifying the probability of making either of these types of
error. Usually the probability of making the first type of error
(incorrectly rejecting the postulated distribution when in fact
it is identical with the true distribution) is specified and no
statement is made with regard to the second type of error. The
specification of the first type of error is expressed as a "signifi­
cance level." Common significance levels are o.oi, 0-05, and
0.10. Thus, when a test is made at the 0-05 (50/o) level, the
engineer takes the chance that 50/, of the rejected postulated
distributions are in fact identical with the corresponding true
distributions. For a complete discussion of the theory under­
lying this and other statistical tests of significance, the reader
is referred to any standard text on statistics.
The technique of performing the X2 significance test is illus­
trated in Example 4- In the table of Example 4 the observed
frequency of wrong connections is shown along with the pos­
tulated theoretical frequency distribution (computed in Ex­
ample i). The problem is to decide whether the observed data
can be construed as a sample coming from the postulated
10 POISSON DISTRIBUTION

theoretical distribution. The computing techniques described

in Appendix G yield a chi square value Of 7.6 with I I degrees of
freedom. (1t should be noted in the table of Example 4 that the
frequencies Of 3 or fewer wrong connections have been com­
bined as have the frequencies of 15 or greater. This is done to
meet the requirement of the X2 test that the theoretical fre­
quency be at least 5 in any group. Thus, this combination re­
sults in I 3 groups, with i 3 - 2
1I degrees of freedom.)

Example 4
TEST OF CONNECTIONS TO WRONG NUMBER (Data from Example i)
Number of
wrong con- Postulated
nections per 0 bserved theoretical f2
period frequency frequency
X f F F
3 6 6-7 5-4
4 11 10.4 i i.6
5 14 i 8.2 io.8
6 22 26.4 i8.3
7 43 33-1 55-9
8 31 36.o 26.7
An -1.2 45.5
zi t- 1.9
10 35 30-7 39.9
11 20 24.3 i6-5
12 i8 17.9 i8.i
13 12 12.0 12.0
14 7 7-5 6-5
15 8 8.6 7-4
TOTAL 76-7 267.0 274.6
X2 - 274.6 267.0
7.6
From statistical tables or from Figure 3 the value of X2 at
the 0-05 (5%) significance level for v
I I is found to be 19.7-*
* In using tables of X2 care should be exercised to note the manner in which
the table is entered with the significance level. If the table is so constructed
that for a given number of degrees of freedom the value of X2 increases with
decreasing percentiles (probabilities), the table is entered with the percentile
corresponding to the significance level. If the table is such that the value of X2
increases with increasing percentiles, the table is entered with the significance
level subtracted from z.
 POISSON AND TRAFFIC 11

This is known as the critical value for the significance test at

the 50/, level. Since the computed value of X2 (7.6) is less than
the critical value, decision 2 is accepted. (The acceptance of
decision 2 is rigorously stated, "There is no evidence to indicate
that the true theoretical distribution differs from the postulated
distribution.") Had the computed value of X2 exceeded the
critical value, decision i would have been accepted.
Once the decision has been made that a particular postulated
theoretical distribution appropriately represents the population
from which the observed data came, the theoretical distribution
can be used in place of the observed distribution for engineer­
ing analysis and action, since it will be free from the random
variations present in the observed distribution.

Interpretation of Examples I to 3
When the engineer gathers field data he is usually interested in
predicting performance for design or administrative purposes.
Fitting a theoretical curve to empirical data is only one step in
the overall objective. The Poisson distribution is of value only
if it permits useful conclusions to be drawn. Below are some
conclusions which can be drawn from each of examples 1 to 3­

Example i:
a. Wrong connections follow a Poisson distribution with the
parameter m - 8.742.
b. Probabilities computed from the Poisson distribution may be
used to predict the occurrence of wrong connections and thus
may serve as a basis for future design considerations. Future
design might, for instance, provide additional facilities to com­
pensate for the capacity consumed by the wrong connections.
A possible alternative might be the provision of some method
of preventing wrong connections.
c. The effectiveness of engineering changes may be evaluated in
terms of a significant change in Poisson parameter.

Example 2:
a. The occurrence of rainstorms in the population considered
follows the Poisson distribution with the parameter m
1.20.
 POISSON DISTRIBUTION

b. The probabilities may serve as a basis for future design.

Consider for instance the following:
Let D, - the damage caused if one storm occurs during a
year,
D2
the damage caused if two storms occur during a
year, . . . . . . . . . . .

D.
the damage caused if n storms occur during a
year,
P(i) the probability of one storm during a year,
P(.-)
the probability of two storms during a year,
. . . . . . . . . . .
P(n)
the probability of n storms during a year.
Then the expected or average damage per year will be
Dav. D, P(i) + D2 P(2) . ....... + D,, P(n).
Example 3:
a. At the location considered and during the time of day and day
of the week considered the arrival of vehicles follows the
Poisson distribution with the parameter m
o.617 for io-
second intervals.
b. The probabilities may be used for design of traffic control
measures. (Examples appearing later in this paper indicate
some specific methods of applying probabilities to traffic de.
sign problerns.)

Traffic Applications of the Poisson Distribution

Prediction of Arrivals from Hourly Volume
For conditions in which the Poisson distributionapplies-i.e.,
under conditions of "free flow"-it is possible to compute the
probability of o, i, 2 . . . . . . . .k vehicles arriving per time
interval of t seconds provided the hourly volume, V, is known:
t length of time interval in seconds
V hourly volume
n number of intervals (per hour)
36oo
t
m average number of vehicles per interval
V Vt

6-oo 36oo
t
 POISSON AND IRAFFIC 13
Then the probability, P(x), that x vehicles will arrive during
any interval is:
X
MXe V' 3600
P (X) = -XI (j-6-oo)e

The hourly frequency, F, of intervals containing x vehicles is:

X Vt
F. = n P(x) 36oo 3600
L-)
t
!1 ( j6oo)
Vt

If the period under consideration is different from an hour, the

36oo in the first parentheses would be replaced by the appro­
priate length of time in seconds. The value V, however, would
still be the hourly volume. Examples 5 and 6 illustrate this.

Example 5
PREDicriON OFARRIVALS (Low Volume)
V = Hourly volume = 37
t = Length of interval = 30 seconds
M Vt 37 - 30 = 0.308

j _6-oo __j_6__oo
e` 0.735
n = 36oo 36oo = I20

t 30

Fo = n (TO mo e- = (120) (1) (1) (0-735) 88.2

Fi = Fo ( m = (88-2) (0.3o8) = 27-9

I)
F2 = F, m = (27.2) (0.3o8) = 4.2

-2) 9

F>2= n - (F.+F,+F2) = i 2o - (88.2 + 2 7.2 + 4-9) 0 .4

Note: The method of computing F, from FO and F2 from F, is dis­
cussed in section on Methods of Computation.
14 POISSON DISTRIBUTION

The following table compares the predicted frequencies with

frequencies observed in a field study (Elliot Avenue West­
bound).9 The predicted frequency in this example corresponds
to the theoretical frequency of the previous examples.
Number of
vehicles
arrivingper Observedt' Predicted
interval frequency frequency
X f. F_
0 88 88.2
1 27 27.2
2 5 4.2
3 0 0.4
120 120.0

Example 6
PREDic-riON oF ARRIVALS (Moderate Volume)
(3o-Second Intervals for 25 Minutes)
V = 418
t = 30 seconds
M = 418 - 30 3.48
36oo

n = 25 - 6o 1500 = 50 (25 minutes, 6o seconds per minute)

.10
jo

e` 0.0308
F. = n e-- = 50 0.308 .54

etc.

In the following table the various predicted frequencies are

compared with frequencies observed in the field (Redondo
Beach Blvd. Eastbound).

9 Based on field data supplied through the courtesy of the Traffic and Lighting
Division, Los Angeles County Road Department.
 POISSON AND TRAFFIC 15
Number of
vehicles
arriving per Observed Predicted
interval frequency frequency
x f. F.
0 0 1.5
1 9 5.4
2 6 9-3
3 9 io.8
4 11 9.4
5 9 6.6
6 5 3.8
7 1 1.9
8 0 1-3
50 50.0
The following table shows the X2 analysis of the data above:
Number
vehicles Predicted
arriving per Observed (theoretical) f2
interval frequency frequency x
x f. F. F.
I 9 6.9 11.7
2 6 9-3 3.9
3 9 io.8 7-5
4 11 9.4 12.9
5 9 6.6 12.3
6 6 7.0 5.1
50 50.0 53.4

fZ2 n 53.4 - 50-0 3-4

F.
v 6 - 2 = 4

2

X.05 9.49 (The symbol X2 is used to denote the critical
value of j2 at the 0-05 (5%) significance level.
The agreement between the predicted and observed data is ac­
ceptable at the 5% significance level.

Testing Goodness of Fit with a Poisson Population

In some locations traffic parameters will vary with time
i6 POISSON DISTRIBUTION

throughout the day (or week or year) while at other locations

the variability may appear to be entirely random. One way,
in which the traffic engineer may test the randomness of his
data is to assume that it follows the Poisson distribution, and
then to test this assumption by the X2 test. If the 2 test fails,
there is strong evidence that the data are non-random. If the
results of the ?? test are acceptable, the data may be assumed
to be random for the purposes of the traffic engineer. It should
be noted, however, that an acceptable X2 test is less conclusive
than one which fails. The rigorous interpretation of an ac­
ceptable test should be, "There is no evidence to indicate non-
randomness of the data."
Example 7 is an analysis of right turns during 300 3-minute
intervals distributed throughout various hours of the day and
various days of the week. While casual inspection of the ob­
served data might lead one to suspect randomness, the test
shows strong evidence of non-randomness.
Example 7
RIGHT TURNS (Westwood at Pico)

Number of cars

7naking rig-rd Observed 7­

I Iteoretical
turns per 3-min. frequency frequency f2
interval f F F
0 14 6.i 32-1
1 30 23-7 38.o
2 sz6 A6.2 28.i
3 68 59.9 77-2
4 43 58.3 31.7
5 43 45.4 40-7
6 30 29.4 3o.6
7 14 I 6.4 12.0
8 10 8.o 12.5
9 6 3.4
10 4 1.3
11 1 12 o.6 6.6 2 i.8
12 1 0-3
13 0 1.
300 300.0 324-7
(The brackets indicate grouping for the X2 test.)
 POISSON AND TRAFFIC 17
i i68
M= 3- 93 e = 0.0203
300

f2 n = 324.7 300-0 = 24.7

Y4 _T_ ­
v = lo - 2 = 8
2 =
X.05 I5-5
Since 24-7 > 15-5, the fit is not acceptable at the 50/, signifi­
cance level.
Some traffic phenomena may be random when observed for an
interval of one length but non-random when observed with an
interval of a different length. Example 8a shows arrival rates
for lo-second intervals. Here the failure of the X2 test (non­
acceptability of fit) indicates non-randomness. Example 8b con­
siders the same arrivals with 3o-second observation intervals;
the assumption of randomness acceptable at the 57, level.

Example 8a

ARRrvAL RATE-iO-SECOND INTERVALS (Durfee Avenue, Northbound)

Number of Observed Theoretical

cars per frequency'O frequency f2

interval f F F

0 139 i2q.6 149.1

1 i28 132.4 123-7
2 55 67-7 44-7
3 25 23.1 27-1
4 10 5"
5 31 3 1.2 7.2 23-5
6 0.
36o 36o.o 368.i
(As before, the brackets indicate grouping for the X2 test)

368
i6- = 1.022 e- = 0.36o

10See footnote 9.
I8 POISSON DISTRIBUTION

f, - n 368.i 36o.o = 8.i

F
5- 2 3

X.0,5 7.81

fit iq ther fore, not accept at the rO/'- lpvpl­

e --- 0 /V

Example 8b

ARRIVAL RATE-30-SECOND INTERVALS (Durfee Avenue, Northbound)

Number of 0 bserved Theoretical

cars per frequency'O frequency f2

interval f F F
0 9 5.6 14.5
1 i6 17.2 14-9
2 30 26-3 34.2
3 22 26.9 i8.o
4 IP 2o.6 17.5
5 10 12.6 7.9
6 3 6-5
7 714 2.8 10.8 i8a
8 3 1.1

9 I 0-4

120 120.0 125.1

m = 368 3.o67 e 0.047

I20

n = 125-I 120.0 5-I

F

v 7 - 2 5

X.05

This fit is acceptable at the 5% significance level.

Probability of Finding a Vacant Parking Space

Certain analyses of parking may be treated by the Poisson

distribution. Example 9 shows the results of a study Of 4 block

10See footnote 9.
 POISSON AND TRAFFIC 19

faces containing 48 one-hour parking spaces. (Loading zones

and short-time parking spaces were not included). Observations
were taken during the hours Of 2 P-M. to 4 P-M- on the days
Monday through Friday. The observation period was divided
into 5-minute intervals. During each 5-minute interval one
observation was made; the exact time within each interval was
randomized. Each observation consisted of an instantaneous
count of the number of vacant spaces. Since the X2 test shows
good agreement between the observed and theoretical data,
there is no ground to doubt that the distribution of vacant
parking spaces follows the Poisson distribution. Assuming the
Poisson distribution to hold, it is possible to compute the
probability of finding a vacant one-hour parking space within
these four block faces.
Example 9
VACANT PARKING SPACES (Westwood Blvd.)

Number of

vacant i-hour

parking 0 bserved Theoretical

spaces per frequency frequency f2

observation f F F

0 29 25-1 33-5
1 42 39-3 44.9
2 21 30-1 14.7
3 I6 I 6.4 15.6
4 7 6.4
5 2 2.0 9. 1 15.8
6 3 12 0.5
7 0 0.2
120 120.0 124-5
i88
'n = -190 = 1.5 7 e-m 0.209

n = I24.5 - 120.0 4.5

F
v 5 3

X.0,5 7.81
20 POISSON DISTRIBUTION

The probability of finding one or more vacant parking spaces,

P (
- 1), is
P ( - 1)
I - P(O) I - e-n
I - 0.209
0-791

Analysis and Comparison of Accident Data

Several writers have discussed the use of the Poisson dis­
tribution in -accident an-aivsiq. The exam-le which follows is
taken from an unpublished paper by Belmont."
Ninety-three road sections each having average traffic less
than 8,ooo vehicles per day and average usage of less than
40,000 vehicle miles per day were combined to give 45 corn­
posite roads each carrying 35,000 to 40,000 vehicle-miles per
day. The accident records of these roads were then compared
using the Poisson distribution as follows:

Example io
SINGLE-CAR ACCIDENTS

Number of
single-car Number of Theoretical
accidents roads number of
1950 observed roads _fX2
X f. F. F.
0 i8 14-5 22.3
1 14 i6.4 12.0
2 7 9.3
3 4 3.5
4 'I 1-n

5 0 13 0.2 .14-1 12.0

6 0 0-04
7 1 0-007
8 0 0.001
45 45 46-3

m 51 I. 133 e__"' = 0.322 45e = I 4.5

45
11 Belmont, D. M., The Use of the Poisson Distribution in the Study of Motor
Vehicle Accidents, Technical memorandum No. B-7, Institute of Transportation
and Traffic Engineering, University of California, June 20, 1952.
 POISSON AND TRAFFIC 21

f.2

F. 46-3 - 45-0 = 1-3 v = 3 - 2 = 1

2
X.O,, 3.84
Although the fit is acceptable at the 5% level, one might be
suspicious of the road section having 7 accidents when the

probability of this occurrence is O'007 or approximately 2 X

45
10 Eliminating this one road results in the following:

Example iob

SINGLE-cAR AcCIDENTS

Number of

single-car Number of Theoretical

accidents roads number of

-'950 observed roads f2

X f. F. F.

0 18 i 6.2 20.0
1 14 16.2 12.1
2 8.1

3 4 12 2-7 i i.6 12.4

4 1 0-7
5 Oj 0.1

44 44-0 44-5

m = 44 I.000 e- = 0.368 44e- i 6.2

44

f.2 = 44.5 - 44.0 0-5 v 3 - 2 I

F.

2
X.0,5 = 3.84

Since there is no evidence (at the 50/, level of significance) that

there is a difference between the observed and the theoretical

data, one may reasonably assume that each of these 44 roads

has the same accident potential. Thus, among road sections

from this group, any section is as likely as any other to have

from zero to four accidents per year; a section which has four
22 POISSON DISTRIBUTION

accidents is no more dangerous than one having no accidents,

the occurrences being simply the operation of chance.
When only two roads or intersections are to be compared as
to accident potential, the following technique1-2 may be used:
a. Compute u,

where U = XI - X2 - I
VXI + X2

XI = larger of observed values

X2 = smaller of observed values

b. Compare the resulting value of u with the appropriate critical
value selected from table. If the computed value of u is less than
the critical value, the two roads or intersections may be con­
sidered to have the same accident potential.

Critical
Significance value of
level u
0.01 2.58
0-05 i.96
o.io

Example I I
COMPARISON OF AccIDENTS AT Two INTERSEcTiONS

Intersection I: 6 accidents during year considered

2: 8 accidents during year considered

xi = 8 X2 = 6
8 - 6 - I
U= o.267

Since the computed value of u in Example I I is less than the

50/, significance level, there is no evidence of a difference in
accident potential between the two intersections.
:12 Hald, A., Statistical Theory with Engineering Applications, John Wiley and
Sons, Inc-, 1952, PP- 725-726.
 POISSON AND TRAFFIC 23

Cumulative Poisson Distribution

In the previous discussion, the probability of exactly o, i, 2,
items per trial (or time interval) has been computed. In many
problems it is desirable to compute such values as the prob­
ability that the number of items (vehicles arriving) per trial is:
a. k or less
b. greater than k
c. less than k
d. k or greater
These probabilities involve the cumulative Poisson distribution
and may be expressed as follows:
P(x < k) = Probability that x < k
k k
= 2; P(X) = 2; e
X=o X=O
k
P(x > k) = i - P(x < k) = i 2; e-
X=O
k- MX

P(x < k) = P(x < k- i) I - e-

X=O X!
k-i
P(x > k) = i - P(x < k) = i - 2; e-
X=o T!
The following examples illustrate the use of the cumulative
Poisson distribution.

Example i 2

LFXT-TURN CONTROL

This example is patterned after one given by Adams.13

A single intersection is to be controlled by a fixed-time signal
having a cycle Of 55 seconds. From one of the legs there is a left-
turning movement amounting to 175 vehicles per hour. The
13 Adams, op. cit.
24 POISSON DISTRIBUTION

layout of the intersection is such that two left-turning vehicles

per cycle can be satisfactorily handled without difficulty,
whereas three or more left-turning vehicles per cycle cause
delays to other traffic. In what percent of the cycles would such
delays occur?

m avg. left turns per cycle

. 55 36oo '5 - 2.67

2 (2.67)z 2.17
P(X > 3) xj e
X=O

= i - [o.o6q + o.x85 + 0.2471

= I - 0.501 = 0.499
Answer: 49.9%

If a special left-turn phase is provided, in what percent of the

cycles will this special feature be unnecessary by virtue of the
fact that there is no left-turning vehicle?
P(o) e-2.67 o.o69
Answer: 6.9%

ExaMDle i 5z

LEFT-TURN STORAGE POCKETS

The California Division of Highways has applied the cumu­

lative Poisson distribution to the design of left-turn storage
pockets. The nockets 2re cle.Rio-ni-il in such a manner th- d-
I---- --- ___- 111aL L11C

number of cars desiring to make a left-turn during any signal

cycle will exceed the pocket capacity only 4% of the time-i.e.,
the probability of greater than a given number of cars is 0-04­
This may be expressed:

k i I Lt Lt
0.04 = I _ I _
!_
36oo) e Y600
X=O

Where: L
number of left-turning cars per hour
t
length of signal cycle in seconds
k
pocket capacity,(number of cars)
 POISSON AND TRAFFIC 25
This relationship may be solved for k by accumulating terms in
the right-hand member until the equation is satisfied.* A more
direct method of computation, however, is to use tables or
charts showing the various values for the cumulative Poisson
distribution. Such a chart is included in the section on Methods
of Computation.
The values obtained are as follows:
REQuiRE-D STORAGE (Number of Vehicles)
Peak hour left- 6o second I20 second
turn movements cycle cycle
100 4 7
200 7 12
300 9 i6
400 12 20
500 14 24
6oo i6 28
700 i8 32
8oo 20 36

The Exponential Distribution

The previous discussions of arrival rate, etc., have dealt with
the distribution of discrete events (arrivals of cars) within a
given time interval. The distribution of gaps (time spacing)
between vehicles is a continuous variable and is exponential in
nature. This distribution may be derived from the Poisson
distributionas follows:
As before, let:

m average number of cars per time interval

t V

36oo

The equation will in general never be precisely satisfied because of discrete

nature of the Poisson distribution. There will be two values of h which nearly
satisfy the relationship-one too large and one too small. It is then a matter of
engineering judgment which to use.
26 POISSON DISTRIBUTION

Where V = hourly volume

t = length of interval in seconds
P(o) = Probability of zero vehicles during an interval t
moe-n - Vt (A)
P(o) = e- = e 3600

If there are no vehicles in a particular interval of length t,

then there will be a gap of at least t seconds between the last
previous vehicle and the next vehicle. -In other words, P(o) is
also the probability of a gap equal to or greater than t seconds.
This may be expressed:
Vt
P(g>t) = e Y60-0 (B)
From this relationship it may be seen that (under conditions
of random flow) the number of gaps greater than any given
value will be distributed according to an exponential curve.
This is known as an exponential distribution.

Example 14
DISTRIBUTION olF GAPS (Arroyo Seco Freeway)
Total. gaps. = 2 114

Total time = I 753 sec­

m = 214 t = 0.122t
1753
P(9 > 0 = e-0.122t

G = (Expected number of gaps > t) = 2I4e-0.1229

Figure I shows a comparison between the observed and
theoretical resultS.14

14When using equation [B] it should be noted that Vt the number of

T6_551
vehicles arriving during an interval, and G, the number of gaps equal to or
greater than some specified value t, are discrete variables. That is, the number
of vehicles or the number of gaps can take on only integral values o, 1, 2 . ......
etc. Probability, P(x) and time, t, on the other hand, are continuous variables
and can take on fractional values. This is illustrated in Figure i, where the
theoretical frequencies are represented by bars and the probability curve is
represented by a continuous dashed curve.
 POISSON AND TRAFFIC 27

Example 15
SAFF GAPS AT SCHOOL CROSSINGS

(This example is based on the report of a joint committee of

the Institute of Traffic Engineers and the International Asso­
ciation of Chiefs of Police.15)
In studying the natural gaps in traffic at school crossings the
following assumptions may be made:
I. The walking speed of a child is 3-5 ft./sec.
2. There must be at least one opportunity of crossing per minute.
(This implies a minimum of 6o opportunities per hour.)
On the basis of these assumptions it is desired to determine the
critical volume for a street of a given width. The critical
volume here refers to that volume above which special measures
will be required for the safety of the child.
There are two approaches to this problem, and care must be
exercised in the selection of the appropriate method. One ap­
proach is that of the number of gaps greater than the time t
required for a child to cross the street. The expected number
of gaps per hour will be V, and the probability (fraction) of
gaps equal to or greater than t is as derived in equation [B].
Thus, the expected number of gaps per hour which are equal to
or greater than t will be
Vt
Ve 3600
The other approach is the number of t-second intervals per
hour which are free of cars. The number of t-second intervals
per hour is 36oo/t. The probability (fraction) of such intervals
free of cars is as given in equation [A]. Thus, the expected
number of t-second intervals per hour which are free of cars
will be
3600 Vi
_e 600
t

15 "Report on Warrants for Traffic Officers at School Intersections," Pro­

ceedings, Institute of Traffic Engineers, vol. i8, 1947, pp. i i8-130.
28 POISSON DISTRIBUTION

When one considers the fact that a long gap may contain
several t-second intervals during which there is opportunity to
cross, this second approach seems more appropriate.

Introducing the second assumption (at least 6o opportunities to

cross per hour),

36oo e - 3600
Vt = 6o
t

where V,, denotes the critical volume.

From the first assumption (walking speed
3-5 ft/sec),

t = ­ D
3-5

where D is the width of the street.

Substituting:

V
D
36oo - 3.5 e - 3600 -3.5 6o
D

V.D 6oD D
e 12,600 = - ­
i 9,6oo 2io

Taking the natural logarithm of both sides:

T7

In
i 2,6oo 2IO

Vc. = _ 1

In D
D 210

Making use of the relationship

In x = log x - log x
log e 0.4343
 POISSON AND TRAFFIC 29

gives
V
= - i 2,6oo i log D
D 0.4343 210

or

V
= - 29,00 (log D - log 2 i o)
D

or

29,000
V
D (2-322 - log D)

This equation has been adopted by the joint committee of the

Institute of Traffic Engineers and the International Association
of Chiefs of Police. Solution of the equation gives the following
values
Width of street (ft.): 25 50 75
Critical volume (v.p.h.): 1072 361 173

Miscellaneous Techniques

Multiple Poisson Distribution

When one group of events follows the Poisson distribution with
the parameter ml and another group of events follows the
Poisson distribution with the parameter M2, the population
formed by combining these groups will follow a Poisson dis­
tribution with the parameter (ml + M2)- (A formal mathe­
matical proof of this relationship is given in section H of the
Appendix.) From the case for combining 2 groups of events it
follows that when k groups of events are combined the result­
ing Poisson distribution has the parameter (ml + M2 + M3 +
............ + m7,). One traffic application of this relation­

ship is the mathematical description of the total arrivals at an

intersection when the arrivals on each leg are known to follow
the Poisson distribution. Example i6 illustrates this application.
30 POISSON DISTRIBUTION

Example i6a
APjaVAL RATE-SOUTHBOUND (Durfee Ave.)16
Number
of cars
arriving Observed Theoretical
per 3o-sec. frequency frequency f2
interval f F F
0 i8 15.0 2i.6
1 32 31.2 32.8
2 28 32.4 24.2
3 20 22-5 17.8
4 13 11.7 14-4
5 7 4-9
6 0 i.6
7 1-9 0-5 7.2 11-3
8 1 0.1
9 0 0.1
120 120.0 122.i

(Brackets indicate grouping for X2 test.)

M. = 2.o8 0-I25
I20

f2F _n = 122.1 I20.0 2.I

v = 6 4
2
X.05 = 9.49
16See footnote 9.
 POISSON AND TRAFFIC 31

Example I6b
ARRIVAL RATE-NORTHBOUND (Durfee Ave.)
Number
of cars

arriving Observed Theoretical

per 3o-sec. frequency frequency f2

interval f F F
0 9 5.6 14-5
1 i6 17.2 14-9
2 30 26-3 34.2
3 22 26.9 I8.o
4 19 2o.6 17-5
5 10 12.6 7.9
6 8 6.5
7 7 2.8
8 3 14 10.8 18.1
9 1 0-4
120 120.0 125.1

(Brackets indicate grouping for the X2 test.)

M, = 368 3-07 e-nn = 0.047

120

f2F _n = 125-1 - 120.0 = 5-I

v 7 - 2 = 5

2
X-05 11.I
32 POISSON DISTRIBUTION

Example 16c
ARRIVAL RATE-NORTHBOUND AND SOUTHBOUND COMBINED
(Durfee Ave.)
Number

of cars

arriving Ut7served Theoretical

per 3o-sec. frequency frequency f2

internal f F F

0 1 0-7
1 4 -14 3.6 13-5 14-5
2 9
3 19 15.9 22-7
4 20 20-4 i9.6
5 A 21.0 12.2
6 19 I 8.o 20.1
7 10 13-2 7.6
8 13 8.5 19.9
9 4 4
2
11 3 9 5 9 5
1.2 8.5
12 1 0.9i

120 120.0 125.1

Mns 6i8 =5.15 M, + ms 3.0 7 + 2.o8 5-15

120

e n' = o.oo6

y f2 n
I25.1 I20.0 5.1

v 8 - 2 6

2
X.05 15.5

Empirical Formulas Based on the Poisson Distribution

Because of differences between theoretical and physical con­

ditions it is frequently necessary to develop empirical formulas
to adequately represent the true behavior of traffic. Where
 POISSON AND TRAFFIC 33
traffic arrives in a random manner, the Poisson distribution
may serve as a building block for such formulas.
Consider the delays at urban stop signs reported by Raff. 17 If a
gap of t seconds is required for crossing the main street, and if
the side street cars arrive at random, the percent of side street
cars which cross without delay will be equal to the percent of
gaps in the main street traffic which are t or greater. From the
equation B this is:
Vt_
iooe 3600

The percent of side street cars delayed will, therefore, be

T60-0
ioo(i -e- Vt )
Raff has found empirically that, because of sluggishness
0 of the
side traffic, the delays should be much greater than the above
value at high volumes on the side street. He has developed the
following empirical formula to represent this situation.
e-2-5v,/3600 e-2VC/3600
P% = 100 I - i - e-2.5V,/3600 0 - e-VII3600

where V, is the side street volume.

When V,
0. this reduces to the original relationship.

Combination of Poisson Distribution and Other Relationships

Several writers have used the Poisson distribution in com­
bination with other relationshipsto form theoretical expressions
of various traffic problems. Although these techniques are
beyond the scope of the present discussion, the reader may wish
to familiarize himself with some of the papers on the subject.
The following partial list of references is suggested as a start­
ing point in the study of these advanced techniques.
Garwood, F., "An Application of the Theory of Probability to the
Operation of Vehicular-Controlled Traffic Signals," journal of
17 Raff, Morton S., A Volume Warrant for Urban Stop Signs, The Eno
Foundation for Highway Traffic Control, 1950.
34 POISSON DISTRIBUTION

The Royal Statistical Society, Supplement, vOl- 7, no. 1, 1940,

pp. 65-77­
Raff, Morton S., "The Distribution of Blocks in an Uncongested
Stream of Automobile Traffic," journal of The American Statis­
tical Association, vOl. 46, no. 253, March 1951, PP. 114-123.
Kendall, David G., "Some Problems in the Theory of Queues,"
journal of the Royal Statistical Society, Series B, vol. 13, no. 2,
1951, PP- 151-173­
Tanner, J. C., "The Delay to Pedestrians Crossing a Road,"
Biometrika, v01- 38, parts 3 and 4, Dec. 1951, PP- 383-392.
Pearcy, T., "Delays in Landing of Air Traffic," journal of The
Royal Aeronautical Society, v01- 52, Dec. 1948, PP- 799-812.
Statistics of the Poisson Distribution
In section F of the Appendix the mean, variance, and stand­
ard deviation of the Poisson distribution are derived. The
results are:
Mean: A. = m
Variance: o-X1 = m
Standard Deviation: a. = A/m-­
For large values of m the section of the probability curve be­
tween I,., - o., and A, + o.,, is approximately 68% of the total
area; the section between pc.,, - 2o..,, and A, + 2o-,, is approxi­
mately 95% of the total area.

Meihuds ui Computation

Review of Procedure
The following is a review of the procedure of computing
probabilities from the Poisson distribution:

i. Determine the parameter m. This parameter is the average

number of occurrences. It may be determined from observed
or assumed data. A trial may consist of an instantaneous ob­
servation, counting events during a time interval, counting
events in a unit area, etc.
Total number of events observed
M Total number of trials or time intervals, etc.
 POISSON AND TRAFFIC 35
2. Once m has been determined, the probability of x events
occurring at any trial (during any time interval) is computed
from the formula

p(x) = m- el
X!

where x! = X(X - 1) (X - 2) .... 3-2-I

Table i lists several sources of tables of e-., e--, P(x), etc.

Figure 2 is a chart of the cumulative Poisson distribution, and
may be used for rapid calculations where accuracy is not im­
portant. A method of making slide rule calculations is indicated
below.

Chart of the Cumulative Poisson Distribution

Figure 2 is a modification of charts by Thorndike18 and

Working.19 The probability P(x -' c) is plotted against m with
c as a parameter. The probability of x equal to or less than c
is simply read from the curve. To obtain P(x c), the prob­
ability of exactly c, values are read from the chart for x c
and for x (c - i). The following relationship is then used:

P(X
C)
P(X -5
C) - P(X -,-- C - 1)

Slide Rule Calculations

The value of eo can be found by means of a log-log-duplex
slide rule; e--T is, of course, the reciprocal of ea,. Having ob­
tained e--T by slide rule or from tables, the individual terms of
the Poisson distribution may be obtained by means of the
following relationship for P(x + i) which is particularly
adapted to slide rule calculations.

P(x) = m el
X!

18 Thorndike, op. cit.

19 Working, Holbrook, A Guide to Utilization of the Binomial and Poisson
Distributions in Industrial Quality Control, Stanford University Press, 1943.
36 POISSON DISTRIBUTION

mx+le mxe M M
P(X+I) =
= - P(X)

(X + 1)!
7!-(X+ 1) X_+ I

Thus, it follows that:

P(o) = e-

P(I) = -M P(o)

P(2) = P(I)
2

P(3) = M P(2)
3
etc.
Table I
AIDS TO COMPUTATION OF POISSON DISTRIBUTION

Type of Aid
and Scope Source

ex, e- Table of Exponentials

Tables x from o.oo to Handbook of Chemistry and
5.5o by o.oi steps. x from Physics
r-o to io.o o.i ste-s Chemical Rubber Pullinhing
Company, Cleveland
Various editions

ax e-a Molina, E. C., "Poissons Exponen­

- Tables
X! tial Binomial Limit," Van Nos-
a' a r, u w.-a- trar.1 V45

lative

a from o.ooi to ioo.o by

various steps

e-E EJ Tables Appendices VI and VII

i! Fry, T. C., "Probability and Its
L from o.i to i.o by steps Engineering Uses, Van Nostrand,
of o.i. E from 1.0 to 20.0 i928
by steps of 1.o

Individual and cumula­

tive

 POISSON AND TRAFFIC 37

Table i (Continued)

Type of Aid Source

and Scope
mx e-z Appendix VI
X! Greenshields, Bruce D.; Shapiro,
Cumulative Tables m Donald; Ericksen, Elroy L.;
from o-1 to i5.o by steps "Traffic Performance at Urban
of O. I Street Intersections," Technical
Report No. i, Yale Bureau of
Highway Traffic, 1947­
Chart Figure 3
Probability that no. cars Adams, William F., "Road Traffic
arriving in a given time Considered as a Random Series,"
will be x or more (or Institution of Civil Engineers
less than x) journal, Nov., 1936, pp. 121-130+­
Chart
Figure 5
Cumulative
Thorndike, Frances, "Applications
P(X !
- C)
of Poissons Probability Summa­
tion," Bell System Technical
journal, vOl. 5, NO. 4, Oct., 1926,
pp. 604-624.
Slide Rule
The Fredrick Post Company

e-w Versalog Slide Rule

x from o.ooi to io.o on x from o.ooi to io.o on 4 scales.

3 or 4 io-inch scales plus Keuffel and Esser Company

normal slide rule fea- Log-Log-Duplex Trig Slide Rule

tures Log-Log-Duplex Decitrig Slide

Rule

x from o-oi to io.o on 3 scales.

ex and e-w Hayashi, Keiichi, "Fiinfstellige

Tables x from o.o to o.1 Tafeln der Kreis- und Hyperbel­

by steps of o.oooi; from funktionen sowie der Funktionen

0.1 to 3-o by steps of ex und e-," Berlin, 1944, Walter

o.ooi; from 3.0 to 6.3 by de Gruyter & Co.

steps of o.oi; from 6.3 to

io by steps of o.i
38 POISSON DISTRIBUTION

220- ­ Gaps on Arroyo Seco Freeway

Lane; 1, 2 to 2:30 P.M. 1.0
October, jg5o

200­
-0.9

180- ­
-0.8

160- ­
e-jWhere 1=0.122t) -0-7

140- -

-o.6

120- -

-0-5
100- ­

-04
8o- ­ 0.

-0.3
Go-- A - Observed frequencies
Theoretical frequencies
Note: Dashcd curve
40- ­ applies only to proba- -0.2
Bell bility scale.

20 -0.1

0
0 5 10 15 20 25 30 35
0
Length of Gap - t (Seconds)

FIGURE I
 Cumulative Probabilitiesfrom the Poisson Distribution
x=5 6 7 8 9 10 L5 2 30
0.9999

0.999

0.99
N
XN
0.9 \111
N 1

> o.8
0.7
o.6 \A\\V
0 0.5 -
0.4 >
0
0-3

.0
0
0.2

14

0.0 I

0.001

0.00011
0.1 0.2 0-3 0-4 o.5 o.6 o.8 i.o 2 3 4 5 6 7 8 9 10
o
Averaae Number of Vehicles per Interval (m)
.o
Modification of charts by F. Thorndike, Bell System technical journal, (October 1926) and H. Working.
A Guide to Utilization of the Binomial and Poisson Distributions, Stanford UniversityPress, 1943.
FIGURE 2
40 POISSON DISTRIBUTION

40
----- ---------
----- -------- ==
----- ---------- --
ool
- ?I -- ------
30 ----- ----------
p, Po,
[Jill 011F -- 7- Ll I

20

- - - - - - - - - -
----------------

- - - - - - - - - - - - - - -

------ ----------------- -

----------------
- -- - - - - - - - - - - --- - - - - -
10

9
-jHj+
7
-_LL
-- f LHTT

3
5 10 15 20 25

Degrees of Freedom

FiGURE 3
 APPENDIX A

Greek Alphabet
Name Lower Case Upper Case
Alpha a A
Beta 18 13
Gamma 'Y r
Delta A
Epsilon E
Zeta z
Eta H
Theta 0
Iota r
Kappa K
Lambda A
Mu M
Nu v N
xi 6
Omicron 0 0
Pi 7r ri
Rho P P
Sigma 0,
Tau 7 T
Upsilon v y
Phi 0 (D
Chi x x
Psi
Omega

41
 APPENDIX B

Fundamental Concepts
a-rnd Cu-7-nul-intations
The subject of permutations may be easily illustrated by the
example of code words. Consider that 3 cards marked A, B, C
are available and are to be used to form as many 3-letter code
words as possible. The result will be:
BC A BC
A<C B ACB
C BAC
B
A BCA
AB CAB
C<B A C BA
In each case there will be 3 choices for the first letter, 2
choices for the second letter, and i choicle for t-e t1l;r A 1Utcr.
From this it follows that:
Pn, = Permutations of n things taken n at a time
= n(n-i) (n-2) .... 3 - 2 - I

= n! = Lctori--l n
P3' = 3 - 2 - I = 6 which checks the empirical result
If there are 5 cards marked A, B, C, D, E and it is desired
that 3-letter code words be formed, there will be 5 choices for
the first letter, 4 choices for the second, and 3 for the third:
P,5 = Permutations Of 5 things taken 3 at a time
= 5 4 3 6o
5 4 3 5! - -5!
2 , 1 2! (5-3)!

42
 APPENDIX B 43
For the general case of the permutations of n things taken m
at a time

p.n n!
(n - m)!

In these code words the order is important, for ABC is different

from ACB. Consider, however, 3-card hands formed from the 5
cards A, K, Q, J, io. If order were important, there would be
6o permutations as follows:

AKQ AQJ Ajio KQJ AKJ AKio AQio Kjio KQio Qjio

AQK AJQ Aiqj KJQ AJK AioK AioQ Kioj KioQ Qi oj

KAQ QAJ JA io QKJ KAJ KAio QAio jKio QKio jQi o

KQA QJA j iOA QJK KJA KioA QioA jioK Qi OK jioQ

QAK JAQ ioAj JKQ JAK ioAK ioAQ i oKj ioKQ ioQj
QKA JQA iqjA JQK JKA ioKA ioQA i qjK ioQK i ojQ

Note that each vertical group is P33 while the whole array is
P,3. In general, however, the order of the cards is unimportant.
It is the particular group of cards that is important (represented
by the io groups above). When order is unimportant the group­
ing is known as a combination.

5 Combinations Of 5 things . Permut. of whole array

C3 taken 3 at a time Permut. within combination

P
5, 6o
Y3
3 10

C. P.- n!
ta P,,n, = (n - m)! m!
44 POISSON DISTRIBUTION

Factorial Zero
For clarity in certain problems it has been found convenient
to define the factorial zero as follows:

(n-i) n!
n

let: n I

then: (i - i)
0 I
Laws of Probability
The following are two important laws concerning prob­
ability:
i. Total Probability
If two events, A and B, are mutually exclusive (if A occurs,
B cannot occur and vice versa) the total probability that one
of these events will occur is:
P(A or B)
P(A + B)
P(A) + P(B)
2. Joint Probability
If two events, A and B, are independent (the occurrence of
one has no influence on the other) the probability that both
will occur together is:

P(AB)
P(A and B)
P(A)P(B)
 APPENDIX C

The Binomial Distribution

Consider a population in which each item may possess one of
two mutually exclusive characteristics (head or tail, good or
bad, o or i, etc.)
Let: p
probability of occurrence of characteristic A
q
(i - P) probability of occurrence of characteristic
B (non-occurrence of A)
Then, by the law of total probability,
P(A or B)
p + q
Suppose that a sample of n items is drawn from the popula­
tion under the following conditions:
a. The size of the population is infinite. (This restriction insures
that withdrawing the sample does not alter the relative pro­
portion of A and B remaining in the population. The same
result may be achieved with a finite population by drawing
one item, replacing, stirring, drawing the next item, etc.)
b. The sample is selected from the population at random.
Then, by the law of joint probability:
the probability that all n items in the sample are A's
pt
the probability of (n-i) A's and i B
pn-lq here the
the probability of (n-2) A's and 2 B's
p,,-2q2 order of
etc. A's and
the probability of iA and (n-i) B's
pqn-1 B's is sig­
the probability of n B's
qn I tnificant
In general, for m A's and (n-m) B's the probability
p- q1­
where m
ol 1, 2, . . . (n- i), n
In drawing the sample, the order in which the A's occur can
take on many possibilities, the number being equal to the com­
binations of n things taken m at a time
Q

45
46 POISSON DISTRIBUTION

For instance, if the number of items in the sample is 5, there

are C31
io ways in which a sample composed Of 3A's and 2B's
may be drawn. Each of these ways will have a probability of
p3q 2. Thus, by the law of total probability, the probability of

3 A's and 2 B's is:

joblq2

or in general:
Q P-q-
Considering the various possibilitiesof o, i, 2, n A's and n,
(n-i) .... i, o B's, the total probability of occurrence of A's
and B's is given by
n
P(A, B) Y- Q P-q n-
M=O

The right hand member is equal to (q + p)n and hence this

relationship is known as the binomial distribution.
n
2; Q P-q- = (P+q)n
M=o
 APPENDIX D

Derivation of the Poisson Distribution

Introduction: The Poisson Distribution is applicable to popu­

lations having the following properties:

a. The probability of occurrence of individuals having a par­

ticular characteristic is low.
b. The characteristic is a discrete variable.

The Poisson Distribution as a Limiting Case of the Binomial

Distribution

Let n = number of items in sample

p = probabilityof occurrenceof a particular characteristic E

q = (i -p) = probability of non-occurrence of character­

istic E

x = number of items in sample having characteristic E.

Then, from the binomial distribution:

P(x) = Cnpxqn- = Cnpx(i -p)--­

X = 0, 1, 2 .... n

Now let:

p be made indefinitely small

n be very large

pn m, where m is finite and not necessarily small.

M
Then: p
n

47
48 POISSON DISTRIBUTION

p(X) = Cn m M X = 0, 1, 2,, n
MO (I n

n! tMY M
!(n-_x)! n ) n

n! fmV M-)
n n n
T!(n-x)! (I ` (I MY

p(x) Fm' I - M)n] n!

P] K n m 2

(n x)l nz I
n)

[A] [B] [C]

Now, if n---> ­

lim P(X) lim [Al [B] [Cl

n---> co n--->

[lim A ] lim B Flim C

n- --> co In-- cD Ln-- - I

A mX

lim A mX

n--+ co X!

B ( - m- n
n)

lim B e- (See appendix E for proof)

n--+ oo

c n!

M\'

(n-x)!nx I -
­
n
 APPENDIX D 49
When n is very large, negligible error is introduced by repre­
senting n! by one term of Stirling's formula. The same state­
ment holds for (n-x)l

C= -0,_rn nn e n

V 27 (n-x) (n-x)n-z e (n-x) I M nx

n

C "2 0 e-- nn

-\/2_7r(n-x)1 (n-x)n-x e-(n-x) 1M nx

n-x nn-x I

e !___ ­

n (n- x), n--- I M

n

e-z I X
n) n-x)n-x m)x
n n

C X]

n X)n-x My

n n

n X
n)" n)Z1

= [C] [C2] [C] [CI]

C, = e-x

Jim C, = e-x

n-> OD

C, = (I - :Y`
50 POISSON DISTRIBUTION

HM C2 = I

n---> oD

Cs =
X
n)"

lim C, =

(See Appendix E)

C4 =
M

n Y

HM Q = I

n--* oD

limc =[limcil Flim

n---> co
c2i
n-- co I Ln--)- oc) I
UM C3
n- oc) I
UM Q

In- oo

le-- i e i

lim P(X)
n--+ co
7T e---m

Since the main bodv of this discussion assump..q the i-vitPnrP nf

the conditions for the Poisson. distribution the above equation
may be written simply:
 APPENDIX E

Derivation of Limit 1 r-n)n (Used in the derivation of the Poisson

n-4 co n distribution)

Let n I
X
)n . 11 I
lim I M lim (I -mx)z; where n = ­
n- co n I X
- --
OD

1)(MX)2
(X
lim I )(MX) +
1
- -- + co

2 (MX)3
(71)( X IX,X
31

liM M + M' (I - X) M'(1 X) (I -!2X) +

I
- ----) OD
X

M2 M3
M - - - - - ­

But expanding e-- in a McLaurin series gives:

+ M2 M3 - ----­

On
lim I- ­
n
n--> co

51
 APPENDIX F

Mean, Variance, and Standard Deviation of

Poisson Distribution

Mean
The mean of any continuous function is obtained by:

f xf (x) dx
f f(x)dx

(The mean may be considered as the distance to the center of

gravity.)

For discrete functions the comparable function defining the

mean is:

A
4W

Y.-"a&j1AjL

."--en dealing with probabilities:

fW = PW

ff(x) dx = i

or
00
2; P(X) = I
X=O

'U = f xp(x)
or0c)

ju = 2;x P(X)
X=O
52
 APPENDIX IT 53
For the Poisson distribution

P(,)

xm0e- '
X=0 M!

o + me- +
2mle--
+ We- ....
2! 3!

M2
Me-M I + M+ ­
1 2!

Mele,
M
By definition, the variance a2may be expressed:

012 f(X) (X - A),

2;f(x)

For the Poisson distribution

2; (X - M) 2 p(X) 2;(X2-2XM+M2) p(X)
q2
ZP(X)

ZX2p(X) - 2M2;XP(X) + M22;P(X)

The last two terms reduce as follows:

-2M2;XP(X) = -2M(M) = -2M2

M2ZP(X) = M2

The first term may be reduced by the following steps:

Zj2p(X) = 2; [X(X- I) + X] P(X) = 2;X(X- 1) P(X) + ZXP(X)

= 2;x(x - 1) P(X) + M

= o +0 + 2M2 e-
n+ 6m' e- + 12M4e- + + M
2 3 4!
54 POISSON DISTRIBUTION

+M+M2+ +M
= M2 e- (i ­
2!

= ml e- (e-) + M = M2 + M

a2 = m2 + M - 9M2 + M2 = M

Siandar -a' Devialzon

By definition the standard deviation, o-, is expressed by

, = N/-7

Thus, for the Poisson distribution

a = V -M
 APPENDIX G

Notes on the X2 Test for Goodness of Fit

Let: f
Observed frequency for any group or interval

F
Computed or theoretical frequency for same group

Then, by definition:

j2 k F,)2
2; (0
i=j Fi
where k number of groups
Expanding:

X2 k f.' 2fi F. F.2

+
Fj Fj Ti-

k k k

2; 2 X f, + 2; F,

But by assumption in the fitting process:

k k
2; fj = 2; Fj = n

where n
total number of observation
So that:

2 k f,2
X n

Either equation (i) or equation (2) may be used for purposes

of computation. Usually (2) will simplify the amount of work
involved.
55
56 POISSON DISTRIBUTION

The value of X2 obtained as above is then compared with the

value from Figure 3 or from tables of X2. Such tables may be
found in any collection of statistical tables, and relate the
values of X2 and significance level with the degrees of freedom.
The number of degrees of freedom, v, may be expressed:20.21
v
G-2
where G
number of groups
For this value of v to be valid, however, it is necessary that the
theoretical number of occurrences in any group be at least 5.
One writer22 further stipulates that the total number of observa­
tions be at least 50- When the number of theoretical occur­
rences in any group is less than 5, the group interval should be
increased. For the lowest and highest groups this may be
accomplished by making these groups "all less than" and "all
greater than" respectively.
2ODixon, W. J., and Massey, F. J., Jr., Introduction to statistical Analysis,
McGraw-Hill, ig5i, pp. Igo-191.
21Yule, G. Udny, and Kendall, M. G., An Introduction to the Theory of
Statistics, 14th edition, Haffner Publishing Company, 1950, P- 475, P- 476­
22Cramer, Harald, Mathematical Methods of Statistics, Princeton University
Press, 1946, P- 435­
 APPENDIX H

Derivation of the Distribution of the Sum of Independent

Poisson Distributions

Consider a population made up of two subpopulations A and B,

each distributed according to the Poisson distribution.
For subpopulation A
X --M
P(X-) = mae a
X.!

For subpopulation B

P(X,) = T-b e-b

Xb-'

If k items occur in a trial from the total population, there may

be a mixture of x,, and xi, as follows:

X. = k; Xb = 0 x. + xb = k
xl = k - i Xb= I X + A = k
X. = k- 2 Xb= 2 x. + xb = k

X, = 2 Xb= k- 2 X + Xb = k
X. = I Xb= k- i x. + xb = k
X. = 0 xb= k x. + xb = k

P(k) =P(x. = k, A 0) + P(x. = k - i, xb = i)

........ + P(X- 1, Xb = k i) + P(x. = o, xb = k)

M_a 7, e-- Mle ---Mb + Ma

k-1
e-- ma Mb
b
M, k-2 e-Ma mb2 e'b
k! o! k-i ! i! (k-2)! 2!

57
58 POISSON DISTRIBUTION

m. e " Mbk-I e b e-M a Mb 7Ce

+ - II (k- I)! + o! k!

P(k) = e-- Ma e--m b M-' + k Mak-1 Mb + k(k- I) M k-2 Mb2

k! k(k- I)! k(k-i) (k-2)! 2!

7- / 7- - -
-l .%
. . . . . . . . .+ nk&- 2 I M Mb" + Mb"

k(k-
3 I(k- I)
T!­

V k (k - I Mk-2 Mbl
P(k) + km k-1 nab +

+ + k Ma Mbk-1 + Mbk

P(k) e-(Ma+ Mb) (Ma + Mb)k

k!

When there are subpopulations A, B . . . . . . .Z, by application

of a similar argument the distribution for the whole population

is found to be

-(M +in +tn)

(M. + Mb + + M.)k e b +

P(k) ­

k!
 The Probability Theory Applied
To Distribution of Vehicles
On Two-lane Highways

ANDRk SCHUHL

Mr. Schuhl is engineer of bridges and roads in the

French Ministry of Public Works. In France these
engineers are graduated from the Polytechnic School
and then from the National School for Bridges and
Roads. Mr. Schuhl is also a graduate of the National
Superior School of Electricity. During the war he
studied hydroelectrical plants in PyrM es and Massif
Central. He joined the Free French Forces through
Spain early in I944 and served with the first Tactical
Air Force of the Seventh Army Group as a French
liaison officer to the American Army. He came to the
United States in I954 with Mission I49 of the O.E.C.E.
to study traffic engineering and traffic control.

A theoretical study of conditions affecting the traffic of vehicles

on a highway requires the constant use of probability theory.
On the one hand, the distribution of vehicles in each lane is
in part a matter of chance. On the other hand, whenever one
studies the behavior of any large number of individuals, in­
evitable departures from the laws which apply to their totality
are found by statistical analysis to follow certain empirical and
relatively stable patterns. These are, in effect, special laws of
probability.
Reprinted with permission from Travaux, January 1955, revised and updated
by the author. Translated from the French.
59
6o POISSON DISTRIBUTION

In particular, Poisson's law, applying to rare events, has thus

far been the chief theoretical instrument for dealing with prob­
lems of vehicular traffic on two or three lane highways, espe­
cially the problem of determining the distribution of such
traffic both in time and in space. This law, as is well known,
assigns the value e-NO (Nor to the probability that there are

n" vehicles in a time interval 0 chosen at random, N being

the average number of vehicles in unit time. The probability
that there are no vehicles in this interval is

e -2V8

While the results yielded by this formula agree well enough

with actual observation when the traffic density is low (a few
dozen cars per hour per lane), they differ widely from reality
when the density is significantly larger. The reasons for this
discrepancy between theoretical prediction and the observed
data are clear enough. But without going into any analysis of
physical causes, it seems obvious that the gap between theory
and actuality can be considerably narrowed. It is the object of
this paper to show how this can be done.
We shall suppose that the entire set of spacings between suc­
cessive vehicles consists of a number of distinct parts or sub­
sets, each having distinct mean values and each obeying some
Poisson-type law. For simplicity we consider just two sub-sets.
Let tho niimhi-r nF per unit timc f0rl ---- sets I
respectively

,yN and (i -,y)N

with mean spacing-values of t, and t2 seconds respectively, and

with t, < t2- We also suppose that the entire set of spacings is
a set of random and independent elements.
Then the probability that there is no vehicle in an interval

0 would be e T
if all spacings were in the first sub-set and e T-
if they were all in the second sub-set.
 PROBABILITY THEORY 6i
Now the respective times covered by the two sub-sets are
7Nt, and i -yNti,

and clearly
i -yNtj = (i -,y) Nt2
Accordingly, the probability that there are no vehicles in an
interval 0 will be given by
0
P(6) = Xytie + (i -,y)Nt2e 0)

with the relation

Nyt, + N(i -,y)t2 = i,

which merely states that the N spacings cover unit time. Before
proceeding further, it must be observed that the first set of
spacings might apply to retarded vehicles which are prevented
from passing by opposing traffic, and the second set to free-
moving vehicles which are able to pass at will. As vehicles
cannot be considered as mere points, two successive vehicles in
the first set must necessarily be separated by a time interval
having a positive lower bound c. On the contrary, free-flowing
vehicles having opportunities to pass may exhibit spacings
equal to zero.
Hence the law of spacings which we use in practice is not
formula (i), but rather the formula

- 6-e - 0

P(0) = Xy(t,-,E)e "' + N(i -Y)12e " (2)

We must now examine the practical usefulness of the laws

embodied in formulas (i) and (2) in providing answers to
various questions arising out of the study of traffic problems.
They are considered in some detail in what follows.

First Question: Probability of a spacing longer than x.

Suppose that the vehicles in our set are represented by points

on a straight line and that p(x)dx is the probability of a spac­
62 POISSON DISMBUTION

ing between x and x + dx. What then is the probability P(O),

that an interval 0 chosen at random on the straight line, con­
tains no points, all choices of 0 on the line being equally
probable?
Now the probability that the initial point A of the interval
AB 0 will be in a spacing lying between x and x + dx is
kxp (x) dx,
k being defined by the condition
M ­
f kxp (x) dx = i,

which simply states that the point A necessarily falls in a spac­

ing which lies between 0 and oo. The value of k is given by

k = I
I =
'VI
X"
f xp (x) dx

where x is the mean value of x.

If the interval AB 0 contains no points, L being the first
point preceding A, and M the first one following A, with
LM x, it follows that x > 0 and that the initial point A lies
between L and P, where LP x -0 >, 0. Now the probability
that the interval AB 0 contains no points is the same as the
probability that its initial point A lies between L and P, and
this probability is found from the above to be

k (x-O xp(x)dx = N(x-O)p(x)dx

x

L A P B
M

FIGURE I
 PROBABILITY THEORY 63
Hence upon integration over all values of x between 0 and co
we have

P(O) N(x - O)p (x) dx N dx f p(x)dx (3)

fa fo
From this we find by two successive differentiations, that

AX) i d' P(x)

N dxl
Using this result, we find by taking the value of P(x) from (i)
that in this case
X
(x) ti+ I -'y 72 (4)
t2

while the total probability of a spacing longer than x is given by

x X

p(x)dx = y e t' + (i -,y)e t2 (5)

If on the other hand we take the value of P(x) furnished by (2),

then

P(x) = 'y e + 1 _'y e (6)

tj-f t2

and
X-e

f p(x)dx = y e + (i -,y)e (7)

It will be seen by examining the graphs of Figure 2 which has

been taken from Statistics with Applications to Highway Traffic
Analyses by Greenshields and Weida, that there is good agree­
ment between the observed data and those furnished by formula
(7). The legend on the diagram gives the actual numerical
values of the various parameters used in this particular case.
(See Appendix, P- 73-)
 50% 100 O'

4- 90

3 8o
0

0 ExperimentaldatafromGreenshields&Weida,
STATISTICS WITu APPLICATIONS To HIGHWAY Ln

20 TRAFFIC ANALYSFS. (Traffic volume Of 550 vc- 70

hicles per hour in one lane)

Theoreticalcurvecorrespondingtotheformula
I- 0
0.56e 0-9) + 0.50e 11-85 V.
7=0-50 tl=1-9 t2=11-85 C=O.q
60
-Theoretical curve corresponding to Poisson's
Law. (N =550)

- 50
0 1 2 3 4 5 6 7 8 gio 15 20 s!5 30 35 40 45 50 55 6o 6i
Value of "t" in Seconds
 PROBABILITY THEORY 65
Figure 3 gives another example, studied on a French road,
showing how actual data are approximated by the theoretical
formula. There were 312 spacings observed in a time period of
52 minutes, 45 seconds. Opposing traffic amounted to only i2
percent of the total flow.

200

150

too'
Observed results
--- Theoretical results
n=n,+ n.
The dotted curves give the di5tribution of
spacings of the first class (constrained vehicles)
and of the second class (free moving vehicles)

50

Vt -------
n: Number of spacings
(cumulated)
0: Length of spacings
in seconds

0
3" 4 5" 6" 7"
I I I
8" 9" 101,
FirURE 3

Second Question: Probability that an interval 0, taken at

random, contains no vehicles, but is bounded by a vehicle
on one side.
First, we clarify the significance of the iterated integrals
occurring in formula W­
66 POISSON DISTRIBUTION

If AB 0 is placed at random on the spacing LM x, AB

can be empty only if A and B are between L and M.

L A B M
U- ;>1<- 0 < -- V
]FIGURE 4

Accordingly, x > 0, and Figure 4 shows that the various cases

in which AB contains no points, may be obtained by letting L
range over all positions preceding A, and M over all positions
following B.
If B lies between M and M', where M1W dv, and if kdv
is the probability of this happening, then the probability that
LM lies between u and u + du is the probability of a spacing
u + 0 + v < x <, u + du + 0 + v,
that is,

p(u + 0 + v)du
Therefore the total probability that L precedes A, M being
given on dv, is

k dv J--p(u + 0 + v)du
0
Also the total probability that L precedes A (u > o) and that M
follows B(v >, o) is

"O k dv fop(u + 0 + v)du

If we make the substitution

U + 0 + V= X,
 PROBABILITY THEORY 67
this can be written in the form

f. - k dv fo+, p(x)dx
By the further substitution
+ V= V,
we find

fo k dy f. P(x)dx = k fe p(x)dx fe dy,

or

k fo (x-O)p(x)dx
where k is determined by the condition that P(O)
I, that is,
k N, and we recover formula (3)­
The upshot of this argument is the demonstration that there
are several distinct definitions of the probability that a given
time interval is empty. In particular our second question can
now be answered, for the probability that a time interval be­
tween 0 and 0 + dO is empty but bounded on one side by a
vehicle, is found from the above results by taking MB v 0,
while L of course precedes A. In fact, if v 0 we obtain from
formula (3) by differentiation

dP d6
TO NdO fo p(x)dx
We shall hereafter write

dP
dO 7(0)

Third Question: Probability that an interval 0, taken at

random, contains exactly n vehicles.
If XI, X2 - ----- X. are the distances of the various points
68 POISSON DISTRIBUTION

(vehicles) interior to 0 measured from the initial point A of the

interval, we can use the preceding results to find the prob­
abilities that the intervals ;Tx-,, XIX2 ------ Xn-1 xn are empty but
bounded on the right by a point (vehicle), namely xi for the
first interval, X2 for the second, and x,, for the last.
Consider the interval W-x,. The probability that it has the
required property is

f 7(xl)dxl
Now the point xi. must be followed by a spacing smaller than
0 -xj, that is, the corresponding interval must be empty but
bounded on the right by a point, ViZ- X2. The probability of this
happening is

f _7(xl)dxlf -7(X2-xl)dx2

We continue in this way, up to the last interval 0 -x, which

is empty but not bounded on the right. Hence the desired
probability is

,7(xl)dxl _7(X2-xl)dX2 ----- 7(xn-xn-,)dxn P(6-Xn)

f f-I

It is this formula which gives the value e -NO (NO)n when

n!
P(O) = e -NI.

Fourth Question: Probable delay in waiting for an empty

interval 0.

This question comes up whenever a vehicle in one lane wishes

to pass another car ahead of it, also whenever a vehicle enters
upon or cuts across a main highway. We first determine the
probability that an interval of length 1, whose initial point is
chosen at random, contains no void of length 0.
This is obviously a function F(l, 0) which steadily decreases
 PROBABILITY THEORY 69
as I increases, 0 being given. The decrease in F(1, 0) due to
an increase dl is the probability that the interval I + dl con­
tains a void 0, while I itself contains no void. It is, in other
words, the probability that the interval I contains a void 0 -dl
at its extremity, dl itself being empty. Consequently it is the
product of the probabilities that 1+ dl -0 contains no void of
length 0 and that 0 -dl is empty but bounded on one side by
a vehicle, provided we neglect all infinitesimals of order higher
than that of F(I -0, 0)_7(0)dl. We can therefore write

= - F(I-0, 0)_70
61
This relation enables us to define F(1, 0) step by step once we
know it on an interval of minimum length 0.
Now
F(1, 0) = i for 0 < 1< 0
and from this we obtain
6F
TT = _ J(O) and F = K-_7(0)1 for 0 < 1 < 20

Since obviously
F(O, 0) P(O),
we see that
K - P(O) + 7(0) -0,
and
F(1,0) -P(O) - 7(0)(1-0)
Continuing in this manner to
nO <, I < (n+i)O
we find that

F(1, 0) = i + 1) (I - no) n-1 + 3n (1-no)n

1(n- i)! n!
We are now in a position to answer the fourth question.
70 POISSON DISTRIBUTION

The probability that the first empty interval of minimum

length 0 begins at a distance 1 from the origin, is equivalent
to the probability that the interval 1 contains no void, while
the interval 0 which immediately follows it is empty but
bounded on one side, and this is
M. O)Y(O)dl.
or, from our definition

SF(1+0, 0) dl.
61
From this it follows that

F(1, O)Y(O)dl = F(O, 0) P(O)

fo
which means, as was to be expected, that the total probability
that a void 0 begins at a positive distance from the origin is the
complement of P(0), which is the probability that the void 0
begins precisely at the origin.
The probable delay in waiting for a void of minimum length
0 can therefore be written as

P IF(1, O)Y(O)dl 13F(1+0, 6) dl

f 61

IF(1+0, 0)11W + fo F(1+0, O)dl

The first term vanishes at both limits, and we also know from
the properties of the function F, that

F(1+0, 0) = - ' BF(1+20, 0)

-7(0) 61
Therefore we can write the second term in the form
P = F(20, 0) i -P(O) 0

T-0) = -,7 (0)

and this is the answer to our question.

 PROBABILITY THEORY 71
It may be noted that this probable value p takes account of
the cases where 1
0, whose probability is P(0).
We must bear in mind that the void 0 required for a passing
maneuver depends upon whether or not the vehicle in question
has to slow down before passing. Taking for example 01 to be
the void necessary for an unretarded passing, and 02 > 01 to
be the void required for a retarded passing, the probable wait­
ing delay is given by

- P(61) 0, P(61)
3(02) P(02)

Fifth Question: Mean number of vehicles contained in a group.

Probability of finding a group with a given number of vehicles.
This question is of interest in the study of the distribution
of spacings. We shall call a collection of vehicles a group if
successive vehicles contained in it are separated by spacings
shorter than a given value t, while successive groups are
separated by spacings longer than t. Now on the basis of the
preceding developments, we know that the probability of a
spacing longer than t is _7(t)
N , and that the probable number of
spacings per unit time is therefore _7(t) where N is the entire
number of spacings.
This is also, as we see, the probable number of groups of
vehicles per unit time, and the probable mean number of
vehicles in a group will be

N
No
Moreover, the probability that there are exactly n vehicles in
a group is

Xt) z Yt) ]n-1

N I N
a result which remains valid for n
i.
72 POISSON DISTRIBUTION

These considerations afford a means of determining the law

of distribution of spacings (by studying the composition of the
various groups) of vehicles which pass a check point.

Conclusion
It appears that the results presented in this paper can lead
to a considerably closer agreement between theory and observa­
tion than has been obtained heretofore. Their interest consists
in the fact that in spite of the somewhat complicated form of
the law we have chosen for the probability of spacings be­
tween successive vehicles, the values which it gives in answer
to our five fundamental questions are comparatively simple.
They involve only the functions P, J, and p. Of course one
could, if one preferred, start with a purely empirical law, as
for example the empirical curve of Figure .2 which is simply
,7(0)
,-, and from this obtain the curves for P and p by graphical
integration or differentiation. Our questions could thus be
answered without any assumed law of probability.
It is hoped that these results will be useful to those interested
in traffic research, who recognize the growing importance of
statistical studies in this field. They may also throw light on
certain extensions of Poisson's Law.
 APPENDIX

For readers interested in the goodness of fit of the time-spacing

distribution represented by the formula here developed, the
following chi-square (X2) test data are submitted.
The author has found it desirable to replace the value of the
various parameters given in the text by the following values:

ti= 1.98 t2= 13A6 'Y = 0.583

e=o.8i I -Y = 0.4,7
These lead to the formula
(O -0.81) 0
0.583e 1.17 + 0-417e 13.16

in percent of the total number of spacings (Fig. I) and

385 X 10-0.37 (0-0,81) +275 XI 0-0.0339
in number of spacings (table).
Col. I gives the limits of the intervals in which the spacings
are classed.

Col. 2 the theoretical number of spacings n, corresponding to

the entries in Column I.

Col- 3 the theoretical number of spacings ft comprised between

times given by Column I.

Col. 4 the observed number f, correspondingto ft of Column 3­

Col- 5 the deviation ft-f,, .

Y t-f,,)2
Col. 6 the values of ft The sum of this column, namely

_f.)
is X2.

73
74 POISSON DISTRIBUTION

FlTTING OF THE LAW

n = 385 X 10-0.37 (0-0-81) + 275 XI 0-0.0339
(Chi-Square Test)

0 n ft f. ft-fl, ft
0 66o
77.66 78 - 0.34 0.00
1 582-34
206-42 207 - 0.58 0.00
2 375-92
97-39 94 + 3.39 0.12
3 278-53
50-19 58 - 7.8i 1.22
4 228-34
29-43 24 + 5.43 1.00
5 198.91
19-97 17 + 2.97 0.44
6 178-94
.15-41 23 7-59 3-74
7 i63-53
8 150-58 12-95 11 + 1-95 0.29

21.80 i8 + 3.8o o.66

10 i28-78
i8.26 23 - 4-74 1.23
12 110-52
15.6 i 20 - 4-39 1.23
14 94-91
i5t.st8 I - 9-69 n L I

i6 81-53 Z)­
i8 70-04 11-49 7 + 4-49 1.75

20 6o.i6 9.88 6 + 3.88 1-52

22 5i.68 8-48 4 + 4-48 2.37

7.28 6 + 1.28 0.22
24 44-40
26 38.13 6.27 6 + 0.27 0.01
12-05 10 + 2-05 0.35
 APPENDIX 75
FITTING OF THE LAW (Continued)
Y _f
)2

0 n ft f. ft-fl, ft
31 26.o8
8-24 11 - 2-76 0.92
36 17-84
5.64 8 - 2-36 0.99
41 12.20
3.86 6 - 2.14 1.19
46 8-34
2.63 1 + x.63 1.01
51 5-71
3-04 4 - o.96 0.30
6i 2.67
1-42 1 + 0.42 0.12
71 1.25
1.25 1 + 0.25 0-05
00 0.00
chi-square 21-24
Degrees of Freedom
25-5
20