COMPAS; modelling of household time constraints

(1)

modelling of household time constraints

August 27

^th

, 2013

Goran Vuk

Vejdirektoratet

(2)

Agenda

1. COMPAS demand model

2. Definition and modelling of PFPT

3. Impact of PFPT to other day pattern models

4. Application and conclusions

(3)

Most of us live in a multi-person household

Nevertheless, the vast majority of regional traffic models focus on individual travelers and thereby neglect family decision-making dynamics and task allocation across the household

Person vs. Household

(4)

COMPAS Demand Model

(5)

Day Level Models

(6)

If a HH has PFPT and duration

Day Level Models

(7)

Mandatory, non-mandatory or home main activity across HH members

Day Level Models

(8)

Number of mandatory activities

Day Level Models

(9)

Day Level Models

E.g. Escorting trips related to work

(10)

Day Level Models

E.g. Escorting trips related to non-work

(11)

E.g. Escorting trips related to non-work

Number of non-mandatory activities

Day Level Models

(12)

Tour & Trip Models

Tour Mode and ToD Model Tour Destination Model

Intermediate Stop Generation Model Intermediate Stop Location Model

Trip Mode Model

Trip Time Model

(13)

Tour & Trip Models

Tour Mode and ToD Model Tour Destination Model

Intermediate Stop Generation Model Intermediate Stop Location Model

Trip Mode Model Trip Time Model

New VOT for Commuter Segment in a non-linear (Gamma) form, based on 2008-2011 TU data.

Other Purpose Segmentes and Income Gropus were scalled relative to OTM 5.3 VoT.

(14)

home

sport

shopping

work

Adult 1

home home

work at home

school

work-based meeting work

Adult 2

sport

school

Child

Case study household

(15)

COMPAS Day Demand Model 1. Day Pattern Models

2. Tour & Trip Models

1. Synthetic Population 2. Long-term- decisions 3. LoS files

Output for application:

1. Trip tables 2. Logsums 3. Other outputs

(16)

Synthetic population information for each person in the household includes:

A) The HH he/she belongs to

B) Person type (7 categories, see next page) C) Age

D) Sex

E) Employment (full time, part time, self employed) F) Work location

G) Student status H) School location

I) Work/school location parking available J) Education level

K) PT-pass ownership L) Has bicycle

Input to the demand model

(17)

Synthetic population information for each household includes:

A) HH size

B) No. of adults C) No. of children D) Car ownership E) HH income F) HH location

G) Type of dwelling

Input to the demand model

(18)

Person types:

1. Full time worker (incl. self-employed) 2. Part time worker

3. Non-working adult 4. Retired

5. Pre-school child

6. Elementary school child

7. Child 16+ (aggregated gymnasium and university)

Input to the demand model

(19)

Outputs from the demand model

The following tables come out of the COMPAS demand model:

1. Household Day 2. Person Day 3. Tour

4. Trip

5. Joint tour

6. Partial joint half tour 7. Full joint half tour

8. Logsums (zone based, many categories for each zone)

(20)

hhno: household number

pno: person number within household tour: tour identification within person day

half: half tour sequence number (1 or 2) within tour tseg: tour segment, i.e. trip number within half tour tsvid: original survey trip number ID

opurp: origin purpose dpurp: destination purpose

oadtyp: trip origin address type (1-home, 2-usual workplace, 3-usual school location, 4-other) dadtyp: trip destination address type

otaz: origin OTM 5.3 zone dtaz: destination OTM 5.3 zone

mode: mode identification (1-walk, 2-bike, 3-sov, 4-hov driver, 5-hov passenger, 6-transit) deptm: departure time

arrtm: arrival time

COMPAS trip table

(21)

Definition and modelling of PFPT

Primary Family Priority Time has been defined as:

- Time spent at home

- All household members must attend (i.e. 2+ HHs) - Activities are child care or social (e.g. dining)

- Minimum length is 20 min.

- (It must be a workday)

(22)

Definition and modelling of PFPT

During PFPT household important decisions are made for the following workday:

- Bringing children to and from school (escorting trips) - Shopping/making dinner

- Usage of the family (only) car

(23)

Definition and modelling of PFPT

Estimation results (PFPT occurred 206 times)

Fi l e PFPT2. F12 Ti t l e Pr i mar y Fami l y Pr i or i t y Ti me Conv er ged Tr ue Obs er v at i ons 644 Fi nal l og ( L) - 221. 4 D. O. F. 14 Rho² ( 0) 0. 504 Rho² ( c ) 0. 451 Es t i mat ed 16 Aug 13

ASC - 1. 33 ( - 3. 3) HH s i z e 3 - 1. 16 ( - 3. 3) HH s i z e 4+ - 1. 48 ( - 3. 7) Pr e- s c hool c hi l dr en 1. 15 ( 3. 6) One adul t + s c hool c hi l dr en 1. 17 ( 3. 0) Two adul t s , bot h wor k i ng 1. 83 ( 4. 3) One adul t has hi gh educ at i on 3. 54 ( 10. 7) HH wi t h one c ar - 0. 458 ( - 1. 6) HH wi t h 2+ c ar s - 1. 030 ( - 2. 2) HH i nc ome 3- 600. 000 0. 619 ( 1. 6) HH i nc ome 6- 900. 000 0. 332 ( 0. 8) HH i nc ome ov er 900. 000 - 0. 123 ( - 0. 3) Wor k - des t . l ogs um 0. 134 ( 1. 6) Home- des t . l ogs um - 0. 0306 ( - 2. 4)

(24)

Definition and modelling of PFPT

Estimation results (PFPT occurred 206 times)

ASC - 1. 33 ( - 3. 3) HH s i z e 3 - 1. 16 ( - 3. 3) HH s i z e 4+ - 1. 48 ( - 3. 7) Pr e- s c hool c hi l dr en 1. 15 ( 3. 6) One adul t + s c hool c hi l dr en 1. 17 ( 3. 0) Two adul t s , bot h wor k i ng 1. 83 ( 4. 3) One adul t has hi gh educ at i on 3. 54 ( 10. 7) HH wi t h one c ar - 0. 458 ( - 1. 6) HH wi t h 2+ c ar s - 1. 030 ( - 2. 2)

(25)

Definition and modelling of PFPT

Estimation results (PFPT occurred 206 times)

(26)

Definition and modelling of PFPT

Estimation results (PFPT occurred 206 times)

(27)

Definition and modelling of PFPT

Estimation results (PFPT occurred 206 times)

(28)

Definition and modelling of PFPT

Estimation results (PFPT occurred 206 times)

(29)

Impact of PFPT to other day pattern models

There are 16 day pattern sub-models included in the COMPAS day demand model

They are put in the hierarchical order

The model for PFPT is placed at the top, i.e. time allocated for family

quality time must be “prioritized” by all household members

(30)

Therefore, impact of PFPT has been estimated in a number of day pattern sub-models, always with the positive sign and statistically significant:

- Household Day Pattern Type sub-models - Person Mandatory Activities sub-models

- Joint Mandatory Half Tour Generation sub-models - Joint Non-Mandatory Tour Generation sub-models - Person Day Pattern sub-models

Impact of PFPT to other day pattern models

(31)

Different person types in the HDAP model Estimate t-value

Mandatory; Full time worker 0.736 2.3

Mandatory; Gymnasium or university student 1.685 2.2

Mandatory; School child 1.456 2.4

Non-Mandatory; Full time worker 0.814 2.2

Non-Mandatory; Retired 2.798 2.5

Non-Mandatory; Non-working adult 2.843 3.4

Non-Mandatory; gymnasium or university student 2.390 2.8

Non-Mandatory; School child 1.363 2.0

Non-Mandatory Pre-school child 0.786 1.2

Work at Home model Estimate t-value

Work at Home 0.263 0.6

Tour types in the Joint Half Tour Generation model Estimate t-value

Partially Joint Paired Half Tours 1.589 3.0

Partially Joint Half Tour 1 1.803 2.9

Partially Joint Half Tour 2 0.535 1.3

Different activity purposes in the Joint Tour

Generation model Estimate t-value

Shopping 1.144 2.3

Work-based Sub-tour Generation model Estimate t-value

Work-based sub-tour 0.995 1.8

Impact of PFPT to other day pattern models

(32)

Impact of PFPT to other models

Impact of PFPT has also been estimated in:

- Tour ToD model, and

- Trip ToD model

(33)

understanding and

modelling of congestion

Application

(34)

understanding and

modelling of congestion

Application on buying a car

(35)

understanding and

modelling of congestion

Application on buying a car

household decides on no. and order of

escorting trips (SFPT)

(36)

understanding and

modelling of congestion

Application

impact of personal characteristics

(education, job type) on buying a car

household decides on no. and order of

escorting trips (SFPT)

(37)

understanding and

modelling of congestion

Application

impact of personal characteristics

(education, job type) personal order of

activity priorities and their duration (PFPT)

on buying a car

household decides on no. and order of

escorting trips (SFPT)

(38)

understanding and

modelling of congestion

Application

impact of personal characteristics

(education, job type) personal order of

activity priorities and their duration (PFPT)

on buying a car

impact of family characteristics

(children, income, car

household decides on no. and order of

escorting trips (SFPT)

(39)

Conclusions

- the concept of PFPT was easily found in the data

- about 60% of 2+ person households could be identified with PFPT - this can be approached through simple choice modelling

- can find variables to indicate higher/lower probability

- a logsum variable from lower levels of day pattern model will provide linkage

- data processing is tricky, but DaySim package is used to support

estimation

(40)

Conclusions

- for more operational purposes more data is essential

- the concept of PFPT needs to de tested for different model structures within the day pattern model:

- should the PFPT-model be placed at the top of the pattern models

- relaxation of the PFPT definition (e.g. duration)

(41)

Conclusions – Draft Runs

ACTUM Sample - Apply

HouseholdDayFileRecords = 801 PersonDayFileRecords = 2.209 TourFileRecords = 2.845 TripFileRecords = 6.973

Tour Rate = 1.3 tour/person Trip Rate = 3.1 trips/person Trips per Tour = 2.4

GCA Population - Apply

HouseholdDayFileRecords = 1.004.823 PersonDayFileRecords = 1.874.031 TourFileRecords = 2.381.216 TripFileRecords = 5.719.859*

Tour Rate = 1.3 tour/person Trip Rate = 3.1 trips/person Trips per Tour = 2.4

* Port-zone traffic and turists are not

included

(42)

Conclusions – Draft Runs

Departure period

Base peak midday night Total

walk 373869 358744 217133 949746 bike 464712 310252 244984 1019948 SOV 501442 401699 310851 1213992 HOVDrive 329049 291822 210724 831595 HOVPass 344719 344634 255857 945210 transit 336235 210961 212172 759368 Total 2350026 1918112 1451721 5719859

Congestion

Percent change

walk 0.31% 0.38% 0.61% 0.40%

Congestion scenario

(43)

Conclusions – Draft Runs

Road pricing scenario

Departure period

Base peak midday night Total

walk 373869 358744 217133 949746 bike 464712 310252 244984 1019948 SOV 501442 401699 310851 1213992 HOVDrive 329049 291822 210724 831595 HOVPass 344719 344634 255857 945210 transit 336235 210961 212172 759368 Total 2350026 1918112 1451721 5719859 Road

Pricing

Scenario

Percent change

walk 0.42% -0.17% 0.78% 0.28%

bike 1.61% 0.29% 2.25% 1.36%

SOV -5.31% -1.80% 0.26% -2.72%

HOVDrive -7.52% -4.06% -1.16% -4.70%

HOVPass 2.17% 0.83% 0.94% 1.35%

transit 2.72% 0.81% 2.98% 2.26%