Spatial Modeling Centre SMC

1
Spatial Modeling Centre (SMC)

• Laboratory for managing micro data bases
• Developing techniques for spatial-temporal
  analysis of such data
• Developing spatial micro simulation models
• Organized as a unit within the department of
  social and economic geography

2
Geographical, longitudinal population and firm
database (ASTRID)

• Longitudinal micro level data for each person
  living in Sweden any time between 1960 and 2005
• Extensive information of some hundred individual
  socio-economic attributes
• Spatial resolution, 100 m for place of living and
  work
• Temporal resolution, one year in most cases, day
  for domestic mobility and migration
• Covers the time period of 1960-2005 for
  demographic attributes and 1985-2005 for
  socioeconomic attributes, i.e. observed
  biographies for a substantial part of a life
  time.
• Based on register data at statistics Sweden
  (including all information from earlier
  computerized censuses)

3
Attributes of individuals and firms in ASTRID

• Person time and place of birth and death,
  parents, immigration, emigration, place of living
  and work, family, education, earnings, employment
  and unemployment, major transfer incomes,
  location, size, use and value of owned property
  etc.
• Firm production, value added, salary cost,
  branch, localized work places, employed etc.

4
Whats the catch? Conditions for using ASTRID

• Only in situ within SMC and its closed network
• Each user responsible for not revealing
  individual data outside laboratory
• Not connecting to other sources of individual
  data
• Only for research giving support (directly or
  indirectly) to the construction of spatial micro
  simulation models of population and labor market
  developments

5
Detailed maps
Visualisation and spatial analyses
6
Stockholm
7
Obs. pop density in Umeå 1995
8
(No Transcript)
9
Observed 3km, 30 km
Population change between 1986 and 1995 (3 km
left, 30 km wright)
10
Observed (2001, left) and simulated (2020, right)
local tax rates as a consequence of aging and
mobility
The areas of the regions are proportional to
the number of inhabitants.
11
A historical study 200 years of population
redistribution (Håkansson 2000)

• The 290 Swedish municipalities of today can be
  compared with the same areas two hundred years
  ago.
• If current population per municipality is
  reduced proportionally (by almost two thirds)
  into the domestic sum of 1800 and compared with
  observed numbers at that time, three quarters of
  the relative distribution remains the same as
  today.
• Thats a result of the very slow speed of
  urbanization that has mainly been a local
  process within current municipality borders.
• So, three quarters of the explanation to the
  contemporary spatial distribution of population
  in Sweden has to be looked for in location
  factors operating before 1800-hundred.

12
But Hidden by the slow urbanization process, is
a dramatic increase in general accessibility

• Two hundred years ago, the average Swede had
  access to 5000 other inhabitants within one hours
  travel by foot.
• Today, the average person in Sweden has access to
  half a million other persons within one hour of
  travel by car, bus or train one hundred times
  more people.
• As a consequence of urbanization, person
  transport developments and general increase of
  domestic population.
• That is probably the single most significant
  prerequisite for the contemporary division of
  labor, specialization and scale in production and
  consumption and hence, for the current level of
  productivity and income.

13

• A micro data based study of agglomeration
  economies, labor mobility and firm performance
• Agglomeration mobility effects of localisation,
  urbanisation and scale on job changes.
  Environment and Planning A (2008). Rikard
  Eriksson, Urban Lindgren Gunnar Malmberg.
• Localised Mobility Clusters Do Networks of
  Flexible Labour Market Relations Affect Firm
  Performance? Urban Lindgren Rikard Eriksson for
  Journal of Economic Geography.
• Related Variety and Labour Mobility. Ron Boschma,
  Rikard Eriksson Urban Lindgren. Paper presented
  at the Max Planck Institute for Economics Jena,
  Germany (2007-10)

14
1. Does localisation, urbanisation and scale
influence magnitude of labour turnover?

• Specialication (Localisation)
• Diversification (Urbanisation , Jacobs
  externalities)
• Internalisation (Scale)
• On intraregional turnover On interregional
  turnover
• Localisation effect () Localisation
  effect (-)
• Urbanisation effect () Urbanisation
  effect (-)
• Scale effect (-) Scale effect (-)
• So, yes locally..

15
2. Does it matter for performance?

• Belonging to identified clusters of high labor
  exchange significantly increases productivity
• Not only mobility of key personnel matters
• Labour market related externalities has twice
  the effect of industrial specialisation
• Traditional factors most important (size, branch
  etc.)
• The positive effect of mobility is not immediate
  its not visible until at least three years of
  employment in the new site.

16
3. How different should new competence be
compared to the existing mix? And what about
related- and unrelated variety?..for giving
positive effects on work place productivity

• All recruitment
• Similar skill portfolio negative effect
• Related skill portfolio positive effect
• Unrelated skill portfolio No significant impact
• Local recruitment Only related skill portfolio
  gives positive effect
• Non local recruitment all skill differences give
  positive effect

17
Creating nine million artificial biographies -
why?

• Core group Einar Holm, Urban Lindgren, Erling
  Lundevaller, Magnus Strömgren, Peter Linder
• At Spatial Modeling Centre (SMC), Social
• economic geography, Umeå University, Sweden

18
Individual Biographies 1
x---------- 11015489 1 1991 34 0 0 ------------ M
aria was born 1991. Mother was Ylva, 29 years old
(510509) Maria started school 1998 2007 Maria
finished school 2007 Maria started work 2012
Maria and Olof (478164) started coabitation 2013
Maria started univ. education 2014 Maria gave
birth to the son Klas (11336182) 2015 Maria
interrupted univ.education ............. Maria
was retired 2056, 65 years old Maria died 2084,
93 years old
19
Individual Biographies 2

• Synthetic biography for 5633246 Siv Peterson
• Siv Peterson was born 1952 in Stockholm. Her
  parents were Johan and Kerstin Peterson. In
  January 1969 Siv Peterson and Arthur Person
  became a couple. At that time they where 15 and
  18 years old respectively. Siv Peterson finished
  school in august 1968. In august 1969 she moved
  to Malmö in order to enter an education within
  the sector of education. In February 1971 she
  separated from Arthur Person. In 1971 Siv
  Peterson and Ivar Larsson became a couple. At
  that time they where 18 and 19 years old. In
  august 1971 they moved together. In august 1975
  Siv Peterson finished her education. She was 23
  years old, cohabiting and had graduated in
  teaching. In august 1975 Siv Peterson didnt find
  any vacant job. In November 1976 Siv Peterson
  gave birth to the daughter Barbro, her father was
  Ivar Larsson. In April 1978 Siv Peterson didnt
  find any vacant job. In September 1980 Siv
  Peterson gave birth to the daughter Siv, the
  father was Ivar Larsson. Siv Peterson was
  employed in august 1981 as a teacher. Siv
  Peterson moved from Ivar Larsson in January 1983.
  Siv Peterson stopped working in February 1987. At
  that time she was 34 years old. Siv Peterson was
  employed in February 1987 in the manufacturing
  sector. After that she changed employment four
  times before getting retired 2018. In august 2033
  Siv Peterson died, 81 years old.

20
Why micro?

• Individuals are the only decision makers in
  society although some possess more power than
  others do. Most social science theory and
  findings are about individuals. Why not use it
  without distortion?
• The way in which individual trajectories interact
  with and constrain each other is often completely
  blurred if they are not described, analyzed, and
  modeled individually.
• Aggregation prior to analysis and modeling of
  trajectories over the state space of individuals
  with several attributes distorts not only
  individual but also aggregate results (honestly,
  this is just a believe statement).

21
Why simulate?

• Compared to i.e.
• Stay with a multivariate analysis
• Construct a macro model
• Stay with pure description
• Do a qualitative study
• Write a novel
• Because its (almost) the only option for making
  controlled experiments in social science.

22
Durkheim-Weber
Criticism of individual-based analysis and
modeling is often based on structuralist views in
social science. They claim that it promotes an
individualist view of the social world, i.e.,
towards methodological individualism. The two
classical sociologists Weber and Durkheim have
come to personalize opposite views in this
debate. Durkheim argues that society is
something entirely different from its individuals
and that its properties cannot be explained by
reference to the properties of the individuals,
while Weber is often cited as saying there is
no society beyond the aggregate of its
individuals.
23
SVERIGE(Sweden in Swedish)
System for Visualizing Economic and Regional
Influences Governing the Environment SVERIGE is
a dynamic and spatial microsimulation model
24
Why use an agent-based model like SVERIGE?

• It enables experiments not possible, not
  economical or not ethical to perform in the real
  population.
• Any aspect of policy, structural conditions and
  behavioral assumptions can be changed.
• Heterogeneity is maintained.
• Consistent results for all scales from micro to
  macro
• The experiment scenario can be compared to a
  consistent reference scenario which differs from
  the experiment only with respect to the changed
  condition and nothing else. The difference in
  outcome is the partial impact.
• It consistently embraces long term dynamic side
  effects like those emerging from future
  interaction between actors.

25
Why use all 9 million individiuals?

• It is possible!
• Its required for high spatial resolution output
• It enables a free choice of scale in drivers and
  outcome
• It facilitates a more accurate modeling of
  competition and cooperation. The partner market,
  labor market, housing market, and education
  market are highly scale-dependent. Anything less
  than full-scale representation truncates the
  choice sets and might force the model to produce
  biased results due to the combinatory of normal
  behavior. The sum of all small, but different
  alternatives is considerable.

26
What is the SVERIGE model?
It is a laboratory a constructed society
containing artificial counterparts to all (or
many) Swedes. In it, they live (and relive) their
lives under controlled conditions. They are
born, grow older, get education, get a partner,
marries, give birth, separates, immigrate,
emigrate, move from home, move locally and
regionally, get job and income, get transfer
payments (for age-retirement, early retirement.
sick leave, unemployment, training, parental
leave and education), they buy consumption goods
and services and finally they die.
27
TG, MS and MAS
Computer Science
Economy and policy Guy Orcutt
Time Geography, Torsten Hägerstrand
Micro simulation (Dynasim, CORSIM, SESIM etc.)
Multi Agent Simulation (ABM, ABSS etc.)
Evsim- eventdriven agent simulator
Turbo time driven many agents simulator
SVERIGE1, 2 och 3
28
How to do micro simulation?
T
T1




Ulf Lundström
Frida Lundell
Klara Pettersson
Klara Pettersson
Lisa Andersson
Lisa Andersson
Per Andersson Age 37
Sex male Birth country Sweden Educ.
level average living place Umeå
Family partner Income 201000 kr
Reg. unempl. 6 Etc
Per Andersson Age 38
Sex male Birth country Sweden Educ.
level average living place Lund
Family single Inkomst 272000 kr
Reg. unempl. 5 Etc
Simsalabim
29
Modules

• Employment and earnings
• Work
• Labour supply
• Retirement
• Early retirement
• Unemployment
• Training
• Parental leave
• Sick leave
• Labor demand

• Demographic modules
• Mortality
• Fertility
• Cohabitation and marriage
• Divorce
• Education

• Mobility
• Emigration
• Immigration,
• Interregional migration
• Residential migration
• Leaving parental home

• Application specific like
• CO2 emision and energy use
• Tourism
• Local economic growth

30
How? By evaluating estimated behavioural
equations for each person each year.
Eg. mortality equation for 60 to 74 year old
persons x -1.2143 - (gender 0.6821) - (age
0.1220) (age2 0.00168) - (EDU2 0.2147) -
(EDU3 0.4137) (DIVSEP 0.2504)
P ex/(1ex)
The calculated death probability (P) is compared
to a random number (0-1). If less, the person
dies. Otherwise the person surlive this year and
continue into the other modules
31
Income and earnings

• For each of eight incomes, two equations are
  estimated
• A decision equation, a probit model producing the
  probability for this person in his/hers current
  situation to work or participate in a program
  next year.
• A magnitude equation, an OLS model predicting the
  amount of income received from that source next
  year for those selected.
• The two equations are connected by help of
  Heckmans two-step procedure in order to
  compensate for selection bias.

32
Destination choice

• Destination choice is calculated as an origin
  constrained interaction model with distance to
  place of work and to all potential destinations
  and populations of destinations as drivers.
• Pij DbijBajDgjk/SjDbijBajDgjk
• where
• Pij prob. for person i to move to j.
• Dij distance between origin for i and
  destination j.
• Bj population in destination j.
• Djk distance between destination j and place
  of work k for person i.

33
33060 labor markets

• For the purpose of modelling labour market
  clearing, demand and supply are divided into 114
  professions times 290 municipalities.

34
How does the employer employ?

• By creating a choice situation with several
  applicants simultaneously competing for a given
  vacancy and then pick one based on a comparison
  of their skills. Or
• By evaluating them one by one and eventually pick
  the last one based on a satisfying rationale, he
  or she is the first one with enough skills for
  the position.

35
Satisfying!

• Behavioral support can be given for both the
  optimizing and the satisfying principle, but
  regardless
• An algorithm leaning closer towards the
  satisfying behavior is the only one efficient
  enough when the population is large, so
• Market clearing procedure Remaining vacancies
  are succesively evaluated against randomly
  selected remaining applicants each year.

36
Market clearing procedure

• The number of demanded positions next year in
  each labour market (professionmunicipality) not
  yet allocated to applicants are contained in a
  vacancy matrix.
• All persons in labour force not yet given work
  next year are contained in a dynamic supply list.
  
• One person in the supply list is chosen at
  random.
• If there is still a vacancy in that persons
  current labour market and if the person is
  employable then, the vacancy goes to this
  person.
• The person is then removed from the supply list
  and the number of vacancies in the persons labour
  market is reduced by one.
• If the person did not get the vacancy he or she
  remains in the supply list as long as his/her
  number of failed matches that year is lower than
  a set maximum.
• In addition, the applicant gets a (small,
  estimated) chance to change profession and/or
  municipality via that method in the supply
  module.
• After that again, a person in the current supply
  list is randomly chosen for matching
  (occasionally the same one as last time).
• The process continues as long as the number of
  remaining vacancies or applicants is above set
  minimum levels respectively and the average
  number of failed matches per remaining applicant
  is below a set level.

37
Example published applications

• Basic regional population scenarios
• Household CO2 production
• Contagious social practice
• Determinants of labor supply
• Immigration effects
• Effects of local firm closure
• China model
• Spatial tourist attraction
• ANN and GA as model estimator
• Diffusion of smallpox
• Aging and taxes
• Effects of locating a repository for nuclear waste

38
Long term effects of large investments on small
and medium sized Swedish towns. -Urban Lindgren-
Analysis of socio- economic consequences of a
large investment of a deep respository plant för
radioactive waste.
39
Hur?
A considerably improved road connecting the
municipality with the larger Stockholm region
Direct plus indirect effect on employment in
Östhammar years 2000- 2060 in 9 different
experiment scenarios.
40
(No Transcript)
41
Labor supply the question
Transfer income or work is there a choice? Do
people consider economic outcome when choosing
between work- and transfer income or between
different transfer programs and if so, what
impact has that on labor supply?
42
Null hypothesis

• Individuals doesnt have a choice or, if they
  have, its not used.
• You are on sick leave because you are sick
• You get unemployment support because you are
  unemployed and cant get a job
• You get support for parental leave because you
  have got a child etc.
• And not at all because that earnings alternative
  just now, for you, gives higher income compared
  to other available sources of income!

43
Method

• Each individuals participation in, and income
  from each of eight sources of income is
  described, modeled and simulated by help of
• Income equations estimated on data from the
  ASTRID database
• experiments with the level of payment performed
  by help of the simulation model SVERIGE.

44
What has impact on labor supply?

• Age, sex, education birth country, employment
• branch, region etc.
• 2. Work- and transfer incomes last year
• 3. Diffussion of changed norms from the
• surrounding region
• Economic conditions level of support
• (this is the experimental variable used)

45
Retirement pensions, agelt65, level 10. Impact
on retirement incomes
Referens
Level 10
46
Main results

• The level of support strongly influences the
  number of participants as well as duration of
  participation and cost for all different welfare
  programs. People do make choices between work and
  different program incomes if they can and that
  choice is largely based on economic rationality.
  
• The potential effect on labor supply of changes
  in level of support in the welfare system varies
  
• Large effect Unemployment support
• Early retirement income
• Medium effect Labor market training support
  Sick retirement support
• Small effect Education support
• Parental leave support

47
Modeling people as a virus or as deliberately
planning agents the human condition