Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.

Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 135

135
APPENDIX C
Theoretical Background of Network Screening
for Proportion Method
This method was first proposed by Heydecker and Wu.119 In Combining Binomial distribution (C1) and Beta distribution
this method, the proportion of collision type (pi) at a site i with (C4) results into unconditional Binomial-Beta distribution,
total crashes of ni and target crash xi is assumed to follow the which can be written as follows
binomial distribution.
n B( + x i , + ni - x i )
h(x i /ni , , ) = i
xi
(C8)
n
f (x i /ni , ) = xi (1 - i )ni - xi , 0 x i ni B(, )
xi
(C1)
i
Using Bayes theorem to combine the prior Beta distribu-
where
n
x ()
is a binomial coefficient defined by tion with site-specific collision data (ni, xi) for each site to
derive the adjusted posterior beta distribution which can be
(nx ) = x !(nn-! x)!
written as
(C2)
-1 (1 - ) - 1
g ( i / , ) = 0 < <1 (C9)
The expected proportion at a site, i, is constant for a given B( , )
site and varies randomly from site to site. Heydecker and Wu119 and are posterior parameters and can be defined as
assumed i to follow Beta distribution, which is defined as
= + xi (C10)
(1 - )
-1 -1
g (/, ) = , 0 < <1 (C3) = + ni - xi (C11)
B(, )
Equation C9 is also a Beta Distribution.
where
For the posterior distribution, the expected value for each
()() site, i, is given by the following equation.
B(, ) = (C4)
( + )
E( i ) = (C12)
where and are the parameters of Beta prior distribution +
and (.) defined as
°
Likewise, the posterior variance is given by
(z ) = 0 e - t t z -1dt (C5)
Also the mean of Beta distribution is given by Var( i ) = (C13)
( + )2 ( + + 1)
A limiting value of proportion is predefined say, p*, for a
E() = (C6)
+ given site and collision type. The pattern score is defined as
where E() is the prior estimate of i. the probability that the expected value of i is greater than p*.
Sites are ranked in descending order of this probability. If the
Variance of beta distribution is given by limiting proportion was selected as the median, m the pat-
tern score can be expressed as:
Var() = (C7)
( + ) ( + + 1)
2
P( i > m ) = 1 - B( m , , ) (C14)

OCR for page 135

136
Parameter Estimation of Then can be estimated as
Beta Prior Distribution
= - (C20)
The parameters and of the Beta distribution can be
expressed in terms of moments (mean and variance) as
shown in equations C15 and C16. The mean and variance
from the observed data are used to estimate and .
To illustrate, suppose there are 1, 2, 3, . . .i, . . .m sites under Posterior Beta Distribution
consideration. i is the proportion of a specific collision type for and Pattern Score
site i, that is i = xij / ni, where xij is the total number of target col-
The median, m, of beta prior distribution is such that
lisions of type, j, during the study period at site i and ni is the
total number of all types of collisions at site i during the same 1
period. The mean proportion of target collisions, j, is given by m g() / , )d = 0.5 (C21)
m
Once and are estimated, m can be estimated using an
ij Microsoft Excel worksheet function.
i =1
j = (C15)
m The posterior parameters, and , can be calculated by
where j is the mean proportion of target collision type j. using equations C10 and C11. The pattern score can be cal-
culated using equation C14.
Similarly, the variance is given by To summarize the above discussion, following is a stepwise
1 m xi 2 - xi 1 m xi
2 procedure to estimate the parameters of beta prior and beta
s2 =
m -1
2 - , n2 (C16) posterior distributions, and thereby the pattern score.
i =1 ni - ni m i =1 ni
For a sufficiently large sample, the sample mean, j , rep- 1. Divide the sites into logical groups. For example, two-
resents the expected value, E(j) and the sample variance, s2, lane rural roads analyzed separately from multilane
represents the population variance, Var(). The variance can roads.
also be expressed as 2. Identify the different types of collisions.
2 3. Find total number of collisions of each type during the
- 2 study period in each site, xi.
s2 = 2 (C17) 4. Find total number of all types of collisions in each site, ni.
+1 5. Calculate the proportion, xi/ni for each site and for each
type of collision of interest.
This can be further simplified as 6. Calculate the mean of the proportions for each collision
type, j .
1
-1 7. Calculate variance using equation C16.
8. Calculate and using equations C19 and C20.
s2 = 2 (C18)
1 9. Estimate the median of Beta prior distribution using
+1
Excel function (m = betainv(0.5, , )).
10. Calculate parameters of posterior Beta distribution as
This gives
= + xi and = + ni xi.
2 - 3 - s2 11. Estimate the pattern score using Excel function as P(i >
= (C19)
s2 m) = 1-betadist (m, , ).