Académique Documents
Professionnel Documents
Culture Documents
1
QNMn
2
QMm
3
QMn
4
QPED
5
EXP
(1SPEED + 2LWm + 3LWn + 4LNm + 5LNn + 6NL + 7SHDW + 8LU + e)
Simplified Model
( SHDW )
MCA = k
2
Qmajor
1
Qminor
2
EXP
(
1
SHDW + e)
Log Log- -Linear Version of the Linear Version of the Log Log- -Linear Version of the Linear Version of the
Model Model Model Model
Ln(MCA) = Ln(k) + 1Ln(QNMm) + 2Ln(QNMn) + 3Ln(QMm) + 4Ln(QMn)
Full Model
+ 5Ln(QPED) + 1(SPEED) + 2(LWm) + 3(LWn) + 4(LNm)
+ 5(LNn) + 6(NL) + 7(SHDW) + 8(LU) + e
Simplified Model
Ln(MCA) = Ln(k) +1Ln(Qmajor) +2Ln(Qminor) +1(SHDW) +e Ln(MCA) = Ln(k) + 1Ln(Qmajor) + 2Ln(Qminor) + 1(SHDW) + e
Statistical Analysis Statistical Analysis Statistical Analysis Statistical Analysis yyyy
The Significance of the models were The Significance of the models were
Univariate and Multivariate Analyses were employed to Univariate and Multivariate Analyses were employed to
develop the models develop the models
The Significance of the models were The Significance of the models were
assessed by: assessed by:
* Th diff i l d d i f ddi * Th diff i l d d i f ddi
Only variables found significant (p<0.05) in univariate Only variables found significant (p<0.05) in univariate
analysis were included in multivariate analysis analysis were included in multivariate analysis develop the models develop the models
* The difference in scaled deviance from adding or * The difference in scaled deviance from adding or
removing the variables removing the variables
analysis were included in multivariate analysis analysis were included in multivariate analysis
* The ratio of scaled deviance to its degrees of freedom * The ratio of scaled deviance to its degrees of freedom
(mean deviance (mean deviance ~~ 1) 1)
* The t * The t- -statistic of parameter estimates at 5% statistic of parameter estimates at 5%
significance level significance level
The Final Models The Final Models The Final Models The Final Models
Full Model (All Junctions) ( )
MCA = 0.01109 QNMm
0.2685
QNMn
0.0515
QMm
0.1036
QMn
0.1263
EXP
(0.01515 SPEED 0.1171 LWm 0.0874 LWn 0.01694 LNm +
5
CTRL
6
SHDW +
7
LU)
where:
5
= 0.0 and 0.0315 for CTRL = 1 and 2, respectively
6
= 0 0 0 02174 and 0 02745 for SHDW= 1 2 and 3 respectively
i i i i
6
0.0, 0.02174 and 0.02745 for SHDW 1, 2 and 3, respectively
7
= 0.0 and 0.01873 for LU = 1 and 2, respectively
Simplified Model (All Junctions)
MCA = 0.0003446 Qmajor
0.5906
Qminor
0.281
EXP
0.0708 SHDW
Q j Q
The Other Junction Groups The Other Junction Groups The Other Junction Groups The Other Junction Groups
Full Model
Three-Legged Non-signalised J unctions ee eggedNo sg a sedJ u ct o s
MCA =0.005294 QNMm
0.2188
QNMn
0.0665
QMm
0.132
QMn
0.1808
EXP
( 0.02279 SPEED 0.0969 LWm 0.0706 LWn 0.00738 LNm
5
SHDW +
6
LU )
where:
5
=0.00, 0.00903 and 0.02099 for SHDW =1, 2 and 3, respectively
6
=0.00 and 0.00755 for LU =1 and 2, respectively
Three-Legged Signalised J unctions
MCA =0.003686QNMm
0.2841
QNMn
0.03934
QMm
0.0734
QMn
0.2586
MCA 0.003686 QNMm QNMn QMm QMn
EXP
( 0.02232 SPEED 0.1293 LWm 0.0848 LWn 0.01532 LNm
5
SHDW +
6
LU )
where:
5
=0.00, 0.01011 and 0.01918 for SHDW =1, 2 and 3, respectively
6
=0.00 and 0.01163 for LU =1 and 2, respectively
Four-Legged Non-signalised J unctions
MCA =0.01193 QNMm
0.28658
QNMn
0.1358
QMm
0.06238
QMn
0.12371
EXP
( 0.00859 SPEED 0.1878 LWm 0.04619 LWn 0.00876 LNm
5
SHDW +
6
LU )
EXP
where:
5
=0.00, 0.00564 and 0.00785 for SHDW =1, 2 and 3, respectively
6
=0.00 and 0.00403 for LU =1 and 2, respectively
Full Model
Four-Legged Signalised J unctions
MCA =0.003706 QNMm
0.273
QNMn
0.0718
QMm
0.0425
QMn
0.2042
EXP
( 0.0246 SPEED 0.0852 LWm 0.0828 LWn 0.01016 LNm
5
SHDW +
6
LU )
EXP
where:
5
=0.00, 0.01373 and 0.02438 for SHDW =1, 2 and 3, respectively
6
=0.00 and 0.00788 for LU =1 and 2, respectively
Non-signalised J unctions
MCA =0.01316 QNMm
0.1597
QNMn
0.0973
QMm
0.1071
QMn
0.1336
EXP
( 0.02418 SPEED 0.0967 LWm 0.0907 LWn 0.01079 LNm
5
SHDW +
6
LU )
EXP
where:
5
=0.00, 0.01809 and 0.0502 for SHDW =1, 2 and 3, respectively
6
=0.00 and 0.01789 for LU =1 and 2, respectively
Signalised J unctions
MCA =0.002822 QNMm
0.3241
QNMn
0.0835
QMm
0.0683
QMn
0.1296
EXP
( 0.02602 SPEED 0.0727 LWm 0.0718 LWn 0.01758 LNm
5
SHDW +
6
LU )
EXP
where:
5
=0.00, 0.01755 and 0.02554 for SHDW =1, 2 and 3, respectively
6
=0.00 and 0.01591 for LU =1 and 2, respectively
Simplified Model
Th l dN i li dJ i Three-legged Non-signalised J unctions
MCA =0.0007581 Qmajor
0.5897
Qminor
0.206
EXP
0.0972 SHDW
Three-leggedSignalisedJ unctions Threelegged Signalised J unctions
MCA =0.000294 Qmajor
0.6184
Qminor
0.263
EXP
0.0791 SHDW
Four-legged Non-signalised J unctions
MCA 0001635Q j
0.4935
Q i
0.2298
EXP
0.0648SHDW
MCA =0.001635 Qmajor
0.4935
Qminor
0.2298
EXP
0.0648 SHDW
Four-legged Signalised J unctions
MCA =0.0001196 Qmajor
0.5756
Qminor
0.4033
EXP
0.0295 SHDW
Q j Q
Non-signalised J unctions
MCA =0.0006039 Qmajor
0.5369
Qminor
0.2869
EXP
0.0864 SHDW
Signalised J unctions
MCA =0.0004693 Qmajor
0.5948
Qminor
0.2411
EXP
0.0589 SHDW
Observed vs Modeled Accidents Observed vs Modeled Accidents Observed vs Modeled Accidents Observed vs Modeled Accidents
15
20
d
e
n
t
s
10
v
e
d
A
c
c
i
d
E
q
u
a
l
i
t
y
L
i
n
e
0
5
O
b
s
e
r
v
0 5 10 15 20
Modeled Accidents
Traffic Flow Traffic Flow Accidents Accidents
Full Model Full Model
Traffic Flow Traffic Flow Accidents Accidents
Full Model Full Model Full Model Full Model Full Model Full Model
50
60
n
t
(
%
)
QNMm
Total
20
30
40
s
I
n
c
r
e
m
e
QNMm
QNMn
QMm
QMn
0
10
20
A
c
c
i
d
e
n
t
0
0 20 40 60 80 100 120
Tr af f ic Flow Incr ement (%)
Traffic Flow Traffic Flow Accidents Accidents
Simplified Model Simplified Model
Traffic Flow Traffic Flow Accidents Accidents
Simplified Model Simplified Model Simplified Model Simplified Model Simplified Model Simplified Model
Total
80
100
e
n
t
(
%
)
Qmajor
Total
40
60
t
s
I
n
c
r
e
m
Qminor
0
20
A
c
c
i
d
e
n
0 20 40 60 80 100 120
Traffic Flow Increment (%)
Traffic Flow Traffic Flow Accidents Accidents
Full Model Full Model
Traffic Flow Traffic Flow Accidents Accidents
Full Model Full Model Full Model Full Model Full Model Full Model
60
80
e
n
t
s
(
%
)
40
e
i
n
A
c
c
i
d
e
0
20
I
n
c
r
e
a
s
e
0 5 10 15 20 25 30
Increase in Approach Speed (km/h)
Lane Width Lane Width - - Accidents Accidents Lane Width Lane Width - - Accidents Accidents
15
20
e
n
t
s
(
%
)
M
a
j
o
r
R
o
a
d
M
in
o
r
R
o
a
d
10
n
i
n
A
c
c
i
d
e
0
5
R
e
d
u
c
t
i
o
n
0 0.5 1
Incr ease in Lane Widt h (m)
Shoulder Width Shoulder Width Accidents Accidents
Full Model Full Model
Shoulder Width Shoulder Width Accidents Accidents
Full Model Full Model Full Model Full Model Full Model Full Model
4
e
n
t
s
(
%
)
S
H
D
W
2
S
H
D
W
3
2
n
i
n
A
c
c
i
d
e
0
R
e
d
u
c
t
i
o
0 0.5 1
Incr ease in Shoulder Widt h (m)
Shoulder Width Shoulder Width Accidents Accidents
Simplified Model Simplified Model
Shoulder Width Shoulder Width Accidents Accidents
Simplified Model Simplified Model Simplified Model Simplified Model Simplified Model Simplified Model
20
e
n
t
s
(
%
)
10
n
i
n
A
c
c
i
d
e
0
R
e
d
u
c
t
i
o
0 0.5 1 1.5 2
Incr ease Shoulder Widt h (m)
Traffic Flows on Signalised and Traffic Flows on Signalised and
Non Non--signalised Junctions Those Reflecting signalised Junctions Those Reflecting
Traffic Flows on Signalised and Traffic Flows on Signalised and
Non Non--signalised Junctions Those Reflecting signalised Junctions Those Reflecting
Junction Safety Junction Safety Junction Safety Junction Safety
Non-signalised
J unction
Signalised J unction
15000
20,000
a
y
b
o
t
h
Motorcycle Accidents =1.0 PIA's per year
J unction
10,000
15,000
v
e
h
i
c
l
e
s
p
e
r
d
a
c
t
i
o
n
s
)
5,000
R
o
a
d
F
l
o
w
(
v
d
i
r
e
c
0
0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 40,000
M
i
n
o
r
R
, , , , , , , ,
Major Road Flow (vehicles per day bot h dir ect ions)
Proposed Junctions with Proposed Junctions with
NN l i l i Non Non- -exclusive exclusive
Motorcycle Lanes Motorcycle Lanes Motorcycle Lanes Motorcycle Lanes
GENERALIZED LINEAR MODEL GENERALIZED LINEAR MODEL
Komponen dalam GLM Komponen dalam GLM
In generalized linear modeling, a statistical model consists of
three components: the systematic component, random
components and link function components and link function
The random component describes the error term or probability
distribution
The systematic component describes the way in which the
explanatory or covariate variables combine together to explain
the variation of response variable. The linear combination of the
explanatory variables is called linear predictor
Link f nction or parameter transformation This f nction links Link function or parameter transformation. This function links
the linear predictor to the random component
1. The random component: Error or probability distribution f(y)
which has a mean which has a mean
2. The systematic component: Linear predictor or linear
i f ti ( ) regression function ()
For n explanatory variables:
n
= i Xi = 0 + 1 X1 + + n Xn
i = 0
where: 0 (sometimes called intercept) and i are parameters
to be estimated; Xi is covariates X1 , X2 , Xn
3. Link function or parameter transformation (g), = g().
This function links the linear predictor (systematic
component) to the mean (random component)
In conventional linear regression analysis, In conventional linear regression analysis,
the assumptions are: the assumptions are: the assumptions are: the assumptions are:
1. The probability distribution of the response variable y is
normal, N (,2), with mean and constant variance 2
2 Th li di t (f i 2. The linear predictor (for n explanatory variables) is:
n
= i xi = 0 + 1 x1 + + n xn
i = 0
3. The link function is identity (i.e. no transformation) y ( )
Model Fitting and Parameter Estimates Model Fitting and Parameter Estimates ode tt g a d a a ete st ates ode tt g a d a a ete st ates
The modelling process may be thought of as one
in which the data: y1, y2, .., yn are matched by a
set of theoretical values: 1 2 n set of theoretical values: 1, 2, .., n
For a good model, the set of must close to For a good model, the set of must close to
the data, y. Thus the are highly patterned,
and therefore easier to understand and interpret
than the y
Model fitting is used to explain the relation
between the response and the explanatory
variables
The process of model fitting involves two
basic decisions: basic decisions:
i. The choice of the relationship between the theoretical
values (s) and the underlying parameters of the
model
ii. The choice of a measure of discrepancy which
defines how close a given set of s is to the data g
The first choice relates s to the systematic
component () of the model, and p () ,
The second is governed by assumptions of the
random component
The first choice generally depends on the The first choice generally depends on the
characteristics of the data under study, and the
data can be drawn fromcertain types of variation, data can be drawn from certain types of variation,
spatial or temporal
Th d ti l t f d l fitti i t The second essential aspect of model fitting is to
minimise a measure of discrepancy between the
observed data and the corresponding fitted observed data and the corresponding fitted
values
Thus the parameters of the model are estimated
by minimising the deviance or maximising the
likelihood or log likelihood of the parameters in
the linear predictor
In generalized linear modeling a measure of In generalized linear modeling, a measure of
discrepancy is called the deviance
This term is expressed as parameter D (y;), s e s e p essed as pa a e e (y;),
which is defined by:
D (y;) = 2 (y;y) 2 (;y) = exact model current model
(y;y) is the maximumlikelihood for an exact fit in which (y;y) is the maximum likelihood for an exact fit in which
the fitted values are exactly equal to the observed data
and (;y) is that of the current model. In order to
minimise deviance, (;y) must be maximised.
In conventional linear regression analysis the deviance is
a well-known residual sum of squares
FITTING ERROR (PROBABILITY) FITTING ERROR (PROBABILITY)
DISTRIBUTIONS DISTRIBUTIONS