Household location in response to changes in transport/accessibility

1
Household location in response to changes in
transport/accessibility a microsimulation
approach using SimDELTA

• Olga Feldman, David Simmonds
• ESRC seminar
• May 2009

2
Presentation structure

• DSC focus
• DELTA
• SimDELTA

3
DSC focus
Property market
Labour market
Product markets
Transport market
4
Background to the DELTA package

• Started in 1995
• Two key characteristics
• Add-on to otherwise free-standing transport
  models
• The model be constructed in terms of processes of
  change.
• Various applications in collaboration with
  transport specialists.

5
Structure of the DELTA model

• The economic model
• The urban land-use model
• The migration model.
• The transport model (to which DELTA is linked)

• The meaning of land use
• Land-use modelling is usually concerned mainly
  with
• households and population
• employment
• building stocks (housing, commercial)
• the interactions between all of these
• and less concerned with land itself.

6
Structure of the DELTA model

• The transport model (to which DELTA is linked)
• The economic model
• The urban land-use model
• The migration model.

7
DELTAa transport model sequence
DELTA
DELTA
DELTA
DELTA
DELTA
Transport model
Transport model
8
Links between main submodels and transport model
9
Accessibility model logsum approach
Ai accessibility of zone i Wj opportunities
at j (jobs by seg or retail floorspace) gij
generalised cost from i to j ?D distribution
coefficient
The logsum approach avoids arbitrary thresholds,
and adding new opportunities or new modes always
improves accessibility or leaves it unchanged
10
Accessibility use in DELTA
11
DELTA dynamics

• Simple sequence of processes within one year
• Complex time-lagged linkages over time
• Model is incremental in one year steps so
• starts from an observed database
• produces an updated database for each forecast
  year.

12
DELTA components within 1 year
13
Linkages within DELTA
14
SimDELTA
15
SimDELTA

• A research project commissioned by DfT
• The overall aim was to develop a new,
  microsimulation - based model of household
  location and related processes of change

Main result

• A new package named SimDELTA
• has been developed and
• Calibrated for the South and West Yorkshire
  SWYSimM model this area is chosen because of the
  existence of the modelling system for this area
  known as SWYSM (South and West Yorkshire
  Strategic Model) DELTA START

16
Microsimulation modelling area2001 LA
17
SWYSimM modelling area
18
Data sources

• 1991 and 2001 UK Census of population
• 1991 Sample of Anonymised Records (SARs)
• National Statistics
• The British Household Panel Survey
• Annual Survey of Hours and Earnings

19
SWYSimM

• Full model microsimulation and DELTA (linked to
  the transport model SWYSM but not the full
  land-use and transport model)
• Microsimulation only microsimulation only runs
  with some DELTA inputs
• Static part 1 to synthesise the initial
  database
• Static part 2 additional variables added
• Dynamic part

20
DELTA components within 1 year
21
DELTA components within 1 year
22
SimDELTA diagram
23
Transport models

• The transport model inputs are taken from the
  models developed for SWYMMS.
• 2 transport models
• a Strategic model with about 90 zones, and
  involving aggregate representation of highway and
  public transport supply, linked to a DELTA
  land-use model to form a full land-use/transport
  interaction model and
• a detailed model, employing 570 zones and
  involving explicit representation of individual
  highway links and public transport services.
• The Detailed Transport Model (DTM) has been used
  as the main source of transport inputs for the
  household location research project (base year
  2000).

24
DTM outputs 2000 and 1991 runs

• Generalised costs for car users are obtained from
  the DTM distance and time outputs, using a cost
  per kilometre to convert distance to money and a
  value of time to convert that cost into time
  units (minutes).
• The generalised cost for public transport users
  is output directly in minutes by the DTM and
  consists of the in-vehicle time, walking and
  waiting time (both weighted), boarding/transfer
  penalties and fares (converted into time using a
  value of time).
• Intrazonal costs have been estimated by as
  two-thirds of the distance, cost etc to the
  nearest neighbour zone.

25
Operation of the model (1)

• DELTA runs to forecast most of 1991-92 changes
  the following DELTA modules have to be run
  Development Model, Development Data Processor,
  Model for Consumer Final Demand Calculations,
  Investment Model, Production and Trade Model, Car
  Ownership Model, Transition and Growth Model,
  Migration Model, Location/Relocation Model,
  Employment Model
• The microsimulation is run to forecast
  household/population changes, relocations, etc
  for 1991-92, taking account of some aspects of
  the DELTA forecasts

26
Static model

• JAVA coded simulated annealing model
• 1991 Census and 1991 SAS data are used in the
  process
• Output from simulated annealing contains all the
  Census variables from the SARS for each member of
  every household, not all needed
• Not defined, not adequately described
  variables had to be assigned values
• Other variables had to be synthetically added,
  particularly
• Household incomes, driving licences

27
Simulated annealing process (1)

• A number of household records for one ward are
  taken at random from the microdata (SARs)
• The characteristics of these households are
  tabulated and the resulting synthetic tables
  compared with real Census tables for this ward,
  and the error (mismatch) is calculated
• Assuming there is a significant error between the
  synthetic and real tables, some of the selected
  households are swapped for an equivalent number
  of household records randomly chosen from the
  microdata, and the error recalculated

28
Simulated annealing process (2)

• If the error has substantially increased, that
  swap is rejected, otherwise the swapped records
  are retained
• Swapping continues until a best fit to the
  synthesized data to the real Census tables is
  reached
• The analogy with the physical process of
  annealing centres on the temperature variable
  which is used to control the swapping of records
  at high temperatures more records are swapped
  than at low temperatures.

29
Additional inputs

• Socio-economic status
• Economic status
• Driving licences
• Wage/Income
• Assigning workplaces for workers in the base year
• Potential cohabitees
• Student households and shares

30
Main outputs from the static model (1)
31
Main outputs from the static model (2)

• Area data area ID, population, households,
  families, jobs taken internally by seg, jobs
  available internally by seg, jobs taken
  externally by seg, accessibility, deprivation
  index, area centroid coordinates (easting,
  nothing), dummy is external?, DELTA zone, DELTA
  area.
• Dwelling data dwelling ID, number of rooms,
  tenure, area code, dwelling type.

32
Main outputs from the static model (3)

• Household data household ID, dwelling ID, area
  ID, student house?, share house?, cars, income,
  tenure, families, preferred tenure, preferred
  number of rooms, mortgage value outstanding,
  mortgage years outstanding, savings, DELTA
  household type
• Person data person ID, household ID, family ID,
  area, age, sex, relation to head of household,
  marital status, seg, ethnicity, economic
  category, work area, driving licence, parent (for
  children), education level, job ID, preferred
  economic status.

33
Main outputs from the static model (4)

• Job data job ID, seg, economic status, economic
  status, area code, current wage, vacant?, person
  ID if occupied

34
Dynamic model processes
35
Processes and their sequence (1)

• Individual demographic and other (mainly
  age-related) changes
• Household changes
• Work-related processes
• Household location
• Microsimulation processes which convert outputs
  from the aggregate (SWYSM) model into microdata
• SWYSM outputs used as inputs to derive
  probabilities within the microsimulation model

36
Processes and their sequence (2)

• Job choice for main earners may occur before
  residential relocation (ie change of job by the
  main earner can lead to household relocation)
• Job choice for others (and possibly for main
  earners) is considered after household relocation
  (actually the next year).
• Job acceptance is influenced by wages offered,
  which in turn reflects labour market conditions.

37
Individual demographics etc

• Ageing
• Survival
• Moving to institution
• Birth/multiple birth
• Redundancy
• Entering labour market/Staying in education
• Leaving/re-entering labour market
• Change job
• Retiring from labour market
• Becoming permanently sick
• Becoming other inactive
• Driving licence losing and acquiring

38
Household changes

• Student households and other shares
• Absence from households
• Separation
• Couple formation/marriage
• Household Division
• Household expenditure
• Housing income
• Obtaining/losing car

39
Work-related processes

• Job supply
• Identifying main earner(s)
• Seeking to change job
• Wages
• Accepting/rejecting job/candidate
• Job and workplace choices

40
Household location

• Housing stock process
• Housing prices or rents
• Household in/out migration
• Whether the household is seeking to move
• Housing tenure choice
• Dwelling choice
• Location choice
• Household location/relocation

41
SWYSM to microsimulation

• Changes in labour demand (conversion of aggregate
  model results for employment into effects on job
  availability and redundancy)
• Changes in housing supply (convert aggregate
  model results for housing development or
  demolition into addition/removal of dwelling
  objects)
• Conversion of SWYSM forecasts of migration from
  the rest of the world into new household objects.

42
Other SWYSM outputs used as probabilities

• Changes in car ownership
• Migration between different parts of the
  microsimulated area
• Migration from the microsimulated area to the
  rest of the world (i.e. leaving the model).

43
SWYSM DTM characteristics (1)

• The base year is 2000
• Three time periods (morning peak, inter-peak,
  evening peak)
• Validated road traffic assignment models for each
  time period, using SATURN software, with
  junctions explicitly modelled on the inter-urban
  network and link based speed/flow relationships
  used for major urban areas

44
Tests
45
New dwellings test

• Increase in dwelling supply inWombell South,
  Barnsley, 1000 extra dwellings in 1993

Distribution of the new arrived residents in 1993
1992 1993 1994 1995 1996 1997 1998
Dwellings 4500 5510 5520 5530 5540 5550 5560
Population 10540 12590 13120 13880 14380 14660 15350
Households 4200 4780 4920 5090 5240 5340 5560
Changes in ward 05CCFX
46
Microsimulation modelling area2001 LA
47
Transport test M18 spur
New junction
48
Artificial test
Total population in the zones with improved
accessibility
Total population in all other zones
49
Conclusions

• We believe that the present project has achieved
  a significant step forward in developing the
  first application of a working dynamic
  microsimulation of household change, location and
  commuting.
• There are many modelling issues in need of
  further work.
• Nevertheless we believe that the model as it
  stands can make a useful contribution to a range
  of other ongoing studies.

50
Acknowledgements

• We are grateful to the former DfT Project
  Officers, Russell Harris and Mo Shahkarami, and
  to members of the project Steering Group for a
  number of helpful discussions during the course
  of the project, and for their patience during the
  delays which the project has incurred.

Full report

• is published the DfT website
• http//www.dft.gov.uk/pgr/economics/rdg/hlm/