open-form discrete choice models - 都市生活学・ネッ...
TRANSCRIPT
-
Open-formdiscretechoicemodels
-
GenealogyofDCMs(again) 26
-
DifferenceofGEVandNon-GEV 27
Multinomial Logit (MNL)
Luce(1959), McFadden(1974) Not consider correlation of choice alternatives (IIA)
Easy and fast estimation High operability (easy evaluation for new additional choice alternative benefit of IIA)
2008/9/20 7 18
MNL2
IIDbusrail
busbus
carcar
XrailUXbusUXcarU
Cjj
i
VViP
expexp
closed-form
Cov(U)
2
22
2
2
2
600
0000
railVbusVcarVcarVcarP
expexpexpexp
Multinomial Probit (MNP)
Thurstone(1927) Consider correlation of choice alternates based on Variance-Covariance matrix
Hard and slow estimation (need calculation of multi-dimensional interrelation depend on N of alternatives')
2008/9/20 7 16
MNP3
busrail
busbus
carcar
XrailUXbusUXcarU
Open-formIdentification
1211
1
21
exp2
1
1
1
J
V V
Jii
i
Jii
JddiP
GEV model (Closed-form) Non-GEV model (Open-form)
Non-GEV model has high power of expression, however parameter estimation cost is high.
-
StructuredCovarianceMNP(1) 28
Proposed new probit type railway route choice model considering overlapping problem. (1993, 1998)
This model applied to practical demand forecasting in real Tokyo network, and it used for decision making of railway policy toward 2015. (2000)
Multinomial Probit with Structured Covariance for Route Choice Behavior, Transportation Research Part B, Vol.31, No.3, pp195-207, 1997.
Prof.Morichi Prof.Yai Prof.Iwakura
-
StructuredCovarianceMNP(2) 29
Tokyo Metropolitan has highly dense railway network ! route overlapping problem
Railway lineabout 130Stationabout 1800Passengers40miliion/dayCong. rate: max over 200%
-
StructuredCovarianceMNP(3) 30
In the overlap network that has correlation between routes, Logit model is susceptible to error by IIA property.
Probit is better ? Difficult to setting covariance matrix for each OD pair
structured covariance by divide into two error Difficult to parameter estimation. (multi-dimensional Integral)
reduce computational time using simulation methods
i = iLength +i
Route
Error of depend on route length
Error of route specific
Ui =Vi +i
The Operations Research Society of Japan
NII-Electronic Library Service
Overlapcorrelation
-
StructuredCovarianceMNP(4) 31
Error of depend on route lengthVariance of route utility increases in proportion to the route length.
Error of depend on route lengthError of route specific
Variance-Covariance structure in Error term
Covariance between routes increases in proportion to the length of route overlap.
Error of route specificindependent of each routecov0
Lr :length of route rLrq :overlap length between route r and q2 :variance of unit length
Simplifyusecov.ratio
Estimate only cov. ratio !
-
StructuredCovarianceMNP(5) 32
Apply to the SCMNL for The 18th master plan for urban railway network in TMA (2000)
ExOomiya to Kanda station
Estimation results
To achieve a high prediction accuracy by the relaxation of route overlapObs 10% in all route
: Shonan-Shinjuku line
: Takasaki line
: Keihin-Tohoku line
: Utsunomiya line
: Yamanote line
(1) Omiya (2) Akabane (3) Ueno (4) Tokyo
Saitama Pref.
Inner Yamanote line
(1)
(2) (3)
(4)
Prediction results
-
Ui =Vi +i + i
MixedLogit Model 33
Mixed Loigt (Train 2000)High flexible structure using two error term.Utility function
Error Component: approximate to any GEV model Random Coefficient: Consider the heterogeneity
dist.: basically assume Normal dist. In the case of normal distribution takes a non-realistic value, it can assume a variety of probability distribution (triangular distribution, cutting normal distribution, lognormal distribution, Rayleigh distribution, etc.).
v dist.: assume any G function IID Gamble (Logit Kernel) MNLany G function (GEV Kernel) NL, PCL, CNL, GNL
-
Ucar = Xcar + +carUbus = Xbus + transittransit +busUrail = X rail + transittransit + rail
ErrorComponent:NL(1) 34
Approximation of Nested Logit (NL)
2008/9/20 7 33
MXL5
222
222
2
2
2
2
22
22
00
00
000000
00
000
transittransit
transittransit
transittransit
transittransit
railtransittransitrailrail
bustransittransitbusbus
carcarcar
XUXUXU
1,0Ntransit
Nested
NL!!
2008/9/20 7 33
MXL5
222
222
2
2
2
2
22
22
00
00
000000
00
000
transittransit
transittransit
transittransit
transittransit
railtransittransitrailrail
bustransittransitbusbus
carcarcar
XUXUXU
1,0Ntransit
Nested
NL!!
IID Gamble Logit
Describe the nest (covariance) using structured . Ex: model choice
CarBus
RailTransit nest
Prail =eVrail+ transittransit
eVcar + eVbus+ transittransit + eVrail+ transittransittransit f transit( )dtransit
Prail =1N
eVrail+ transittranistN
eVcar + eVbus+ transittranistN
+ eVrail+ transittranistN
N
Normal nest
Choice prob.(open-form)
Choice prob.(Simulated)
-
ErrorComponent:NL(2) 35
Approximation of Nested Logit (NL)
2008/9/20 7 33
MXL5
222
222
2
2
2
2
22
22
00
00
000000
00
000
transittransit
transittransit
transittransit
transittransit
railtransittransitrailrail
bustransittransitbusbus
carcarcar
XUXUXU
1,0Ntransit
Nested
NL!!
Note that variance-covariance matrix is inconsistent with normal NL
CarBus
RailTransit nest
0 0 00 transit
2 transit2
0 transit2 transit
2
!
"
####
$
%
&&&&
+
2 0 00 2 00 0 2
!
"
####
$
%
&&&&=
2 0 00 transit
2 + 2 transit2
0 transit2 transit
2 + 2
!
"
####
$
%
&&&&
2 0 00 2 transit
2
0 transit2 2
!
"
####
$
%
&&&&
v
Normal NL
Approximated NL based on MXL
Diagonal elements (variance of Bus and Rail) is bigger than transit
-
road2 road
2 0
road2 road
2 + transit2 transit
2
0 transit2 transit
2
!
"
####
$
%
&&&&
+
2 0 00 2 00 0 2
!
"
####
$
%
&&&&
=
road2 + 2 road
2 0
road2 road
2 + transit2 + 2 transit
2
0 transit2 transit
2 + 2
!
"
####
$
%
&&&&
ErrorComponent:CNL 36
Approximation of Cross Nested Logit (CNL)
2008/9/20 7 34
MXL6
222
22222
222
2
2
2
22
2222
22
0
0
000000
0
0
transittransit
transitroadtransitroad
roadroad
transittransit
transitroadtransitroad
roadroad
railtransittransitrailrail
busroadroadtransittransitbusbus
carroadroadcarcar
XUXUXU
1,0, Nroadtransit
Cross-Nested
CNL!!2008/9/20 7 34
MXL6
222
22222
222
2
2
2
22
2222
22
0
0
000000
0
0
transittransit
transitroadtransitroad
roadroad
transittransit
transitroadtransitroad
roadroad
railtransittransitrailrail
busroadroadtransittransitbusbus
carroadroadcarcar
XUXUXU
1,0, Nroadtransit
Cross-Nested
CNL!!
v
2008/9/20 7 33
MXL5
222
222
2
2
2
2
22
22
00
00
000000
00
000
transittransit
transittransit
transittransit
transittransit
railtransittransitrailrail
bustransittransitbusbus
carcarcar
XUXUXU
1,0Ntransit
Nested
NL!!
CarBus
RailTransit nest
Road nest
Describe the nest (covariance) using structured .
-
Uzorn1 = X zorn1 + 1 + 2 + zorn1Uzorn2 = X zorn2 + 1 + 2 + 3 + zorn2Uzorn3 = X zorn3 + 2 + 3 + 4 + zorn3Uzorn4 = X zorn4 + 3 + 4 + zorn4
ErrorComponent:SCL 37
Approximation of Spatial Correlation Logit
2008/9/20 7 35
MXL7
220
20
20
220
20
20
220
20
20
220
2
2
2
2
20
20
20
20
20
20
20
20
20
20
0020
0200
000000000000
0020
0200
43044
3302033
2201022
11011
zonezonezone
zonezonezone
zonezonezone
zonezonezone
XUXUXUXU
1,0,, 321 N
O
2008/9/20 7 35
MXL7
220
20
20
220
20
20
220
20
20
220
2
2
2
2
20
20
20
20
20
20
20
20
20
20
0020
0200
000000000000
0020
0200
43044
3302033
2201022
11011
zonezonezone
zonezonezone
zonezonezone
zonezonezone
XUXUXUXU
1,0,, 321 N
O
2 02 0
2 0
02 3 0
2 02 0
0 02 3 0
2 02
0 0 02 2 0
2
!
"
######
$
%
&&&&&&
+
2 0 0 00 2 0 00 0 2 00 0 0 2
!
"
#####
$
%
&&&&&
=
02 + 2 0
2 0 0
02 2 0
2 + 2 02 0
0 02 2 0
2 + 2 02
0 0 02 0
2 + 2
!
"
######
$
%
&&&&&&
v
Describe the spatial correlation using structured .
zorn
1 zorn
2 zorn
3 zorn
4
-
ErrorComponent:HL 38
Approximation of heteroscedastic LogitAssume the different error variance in each alternatives'Identification problem occur in the case of not fixed one of the parameters to zero at least.
2008/9/20 7 37
MXL9
22
22
22
2
2
2
2
2
2
000000
000000
000000
rail
bus
car
rail
bus
car
railrailrailrailrail
busbusbusbusbus
carcarcarcarcar
XUXUXU
1,0,, Nrailbuscar
Identification0
2008/9/20 7 37
MXL9
22
22
22
2
2
2
2
2
2
000000
000000
000000
rail
bus
car
rail
bus
car
railrailrailrailrail
busbusbusbusbus
carcarcarcarcar
XUXUXU
1,0,, Nrailbuscar
Identification0
car2 + 2 0 0
0 bus2 + 2 0
0 0 rail2 + 2
!
"
####
$
%
&&&&
Rail: High travel time reliabilityError variance is small
consider only heteroscedasitcIID assumption is not relaxed
Car: Low travel time reliabilityError variance is large
Assume heteroscedastic in error bus2 =1FIX !
-
2008/9/20 7 38
MXL10
2,,
21
20
2,
1,,
11
10
1,
,,2,10,
,,110,
*1***
Groupncarncar
GroupGroupGroupncar
Groupncarncar
GroupGroupGroupncar
ncarncarnncarnncar
ncarncarnncar
TU
TU
TmaleTmaleU
TmaleU
21
20
11
10
,
,GroupGroup
GroupGroup
1
0 0+ 1
2 1Group1Group2
ncarncarnncar TU ,,,
ncarncarncar TU ,,,
RandomCoefficient(1) 39
Taste heterogeneity of decision makerParameters defined homogeneously in population. However, decision maker n has different taste ( = heterogeneity)
2008/9/20 7 38
MXL10
2,,
21
20
2,
1,,
11
10
1,
,,2,10,
,,110,
*1***
Groupncarncar
GroupGroupGroupncar
Groupncarncar
GroupGroupGroupncar
ncarncarnncarnncar
ncarncarnncar
TU
TU
TmaleTmaleU
TmaleU
21
20
11
10
,
,GroupGroup
GroupGroup
1
0 0+ 1
2 1Group1Group2
ncarncarnncar TU ,,,
ncarncarncar TU ,,,
2008/9/20 7 38
MXL10
2,,
21
20
2,
1,,
11
10
1,
,,2,10,
,,110,
*1***
Groupncarncar
GroupGroupGroupncar
Groupncarncar
GroupGroupGroupncar
ncarncarnncarnncar
ncarncarnncar
TU
TU
TmaleTmaleU
TmaleU
21
20
11
10
,
,GroupGroup
GroupGroup
1
0 0+ 1
2 1Group1Group2
ncarncarnncar TU ,,,
ncarncarncar TU ,,,
Segmentation (observable heterogeneity) Constant by gender
parameter by gender
2008/9/20 7 38
MXL10
2,,
21
20
2,
1,,
11
10
1,
,,2,10,
,,110,
*1***
Groupncarncar
GroupGroupGroupncar
Groupncarncar
GroupGroupGroupncar
ncarncarnncarnncar
ncarncarnncar
TU
TU
TmaleTmaleU
TmaleU
21
20
11
10
,
,GroupGroup
GroupGroup
1
0 0+ 1
2 1Group1Group2
ncarncarnncar TU ,,,
ncarncarncar TU ,,,
Females constant; 0
males constant:0+1
Females parameter: 2
males parameter:1
1 - malen = femalen
-
RandomCoefficient(2) 40
Parameter distribution (unobservable heterogeneity)Assume the heterogeneity of parameter In the case of parameter following Normal dist., we estimate the dist.s hyper-parameter (mean and variance).
2008/9/20 7 39
MXL11
nrailnrailnnrailnrail
nbusnbusnnbusnbus
ncarncarnncarncar
TTU
TTU
TTU
,,,,
,,,,
,,,,
ncarncarnncar TU ,,,
1,0Nn
2
2,Nn
parameterunknown:,
nn income10
2008/9/20 7 39
MXL11
nrailnrailnnrailnrail
nbusnbusnnbusnbus
ncarncarnncarncar
TTU
TTU
TTU
,,,,
,,,,
,,,,
ncarncarnncar TU ,,,
1,0Nn
2
2,Nn
parameterunknown:,
nn income10
2008/9/20 7 39
MXL11
nrailnrailnnrailnrail
nbusnbusnnbusnbus
ncarncarnncarncar
TTU
TTU
TTU
,,,,
,,,,
,,,,
ncarncarnncar TU ,,,
1,0Nn
2
2,Nn
parameterunknown:,
nn income10
2008/9/20 7 39
MXL11
nrailnrailnnrailnrail
nbusnbusnnbusnbus
ncarncarnncarncar
TTU
TTU
TTU
,,,,
,,,,
,,,,
ncarncarnncar TU ,,,
1,0Nn
2
2,Nn
parameterunknown:,
nn income10 Hyper-parameter can describe using observable variables
2008/9/20 7 39
MXL11
nrailnrailnnrailnrail
nbusnbusnnbusnbus
ncarncarnncarncar
TTU
TTU
TTU
,,,,
,,,,
,,,,
ncarncarnncar TU ,,,
1,0Nn
2
2,Nn
parameterunknown:,
nn income10
depend on observable income variable2008/9/20 7 39
MXL11
nrailnrailnnrailnrail
nbusnbusnnbusnbus
ncarncarnncarncar
TTU
TTU
TTU
,,,,
,,,,
,,,,
ncarncarnncar TU ,,,
1,0Nn
2
2,Nn
parameterunknown:,
nn income10
-
Summaryofopen-formmodels 41
Strengths
v Describe correlation between alternatives by EC MNP: all alternatives (relax and reduce by structuring) MXL: depend on approximated model
v Describe heterogeneity by RC Segmentation, parameter distribution
Limitations
v High calculation cost in parameter estimation Open-form model has high dimensional integration. Recently, proposed high speed estimation methodsEx: Bayesian estimation (MCMC) see Trains book
MACML: analytical integration by Bhat et al.(2011)
-
Reference 42
Yai,T.,Iwakura,S.,&Morichi,S.:Multinomialprobit withstructuredcovarianceforroutechoicebehavior.TransportationResearchPartB:Methodological,31(3),pp.195-207,1997.
Morichi,S.,Iwakura,S.,Morishige,S.,Itoh,M.,&Hayasaki,S.:Tokyometropolitanrailnetworklong-rangeplanforthe21stcentury.TransportationResearchBoard,(01-0475),2001.
Train,K.E.:Discretechoicemethodswithsimulation.Cambridgeuniversitypress,2009.
Revelt,D.,&Train,K.:.Mixed logitwithrepeatedchoices:households'choicesofapplianceefficiencylevel.Reviewofeconomicsandstatistics,80(4),pp.647-657,1998.
Ben-Akiva,M.,Bolduc,D.,&Walker,J.:Specification,identificationandestimationofthelogitkernel(orcontinuousmixedlogit)model.DepartmentofCivilEngineeringManuscript,MIT,2001.
Walker,J.L.,Ben-Akiva,M.,&Bolduc,D.:Identificationofparametersinnormalerrorcomponentlogit-mixture(NECLM)models.JournalofAppliedEconometrics,22(6),pp.1095-1125,2007.