MicMac Combining micro and macro approaches in demographic forecasting A study commissioned by the European Commission 6th Framework Programme Call for tenders: FP6-2003-SSP-3 (May 2005

1
MicMacCombining micro and macro approaches in
demographic forecasting A study commissioned
by the European Commission6th Framework
ProgrammeCall for tenders FP6-2003-SSP-3(May
2005 April 2009) Introduction to the MicMac
project
QMSS2 Immigration and Population Dynamics Leeds,
2 9 July 2009
2
The project
3
Aim of MicMac

• To develop a methodology that complements
• conventional population projections by age and
  sex (aggregate projections of cohorts, Mac)
• with
• projections of the way people live their lives
  (projections of individual cohort members, Mic)

4
Expected outcome of MicMac

• A model and software program to generate
• detailed demographic projections
• that can be used in the context of the
  development of
• sustainable (elderly) health care and pension
  systems

5

Participating institutes

• Consortium
• NIDI - Netherlands Interdisciplinary
  Demographic Institute
• VID - Vienna Institute of Demography
• INED - Institut National dÉtudes Démographiques
• BU - Bocconi University
• EMC - Erasmus Medical Centre
• MPIDR - Max Planck Institute for Demographic
  Research
• IIASA - International Institute for Applied
  Systems Analysis
• UROS - University of Rostock
• Period May 1, 2005 April 30, 2009

6
WP 0 Coordination
NIDI
The Work Packages
Expert Meeting on Assumptions
EMC/UROS
WP 4 Health
NIDI
IIASA
WP 1 Multi-State Methods Education
WP 2 Micro Simulation
WP 3 Uncertainty
NIDI/MPIDR
VID
WP 5 Fertility and living arrangements
NIDI/MPIDR
BU/VID/INED
WP 6 Dissemination of results
7
The model
8
MicMacBiographic forecasting

• A macro-model (MAC)
• Extends the cohort-component model to multistate
  populations
• Cohort biographies
• A micro-model (MIC) that models demographic
  events at the individual level
• a dynamic micro-simulation model that predicts
  life transitions at the individual level
• Individual biographies
• Point of departure LifePaths (Statistics Canada

9
The micro-macro link in demographic
projection The dual approach adopted in the
workplan
   
   
 
Inspired by Coleman (1991) Foundations of social
theory. Belknap Press of Harvard
10
The projection model is a multistate probability
model

• States (attributes)
• At the individual level
• State probability probability that an individual
  has a given attribute at a given age (is in a
  given state at a given age) (state probability)
• At the aggregate (population) level counts
• State occupancy expected value of the number of
  people of a given age with a given attribute
• Transitions between states
• Transition probability transitions / risk set
• Transition rate transitions / exposure time

11

• State variables and covariates
• age
• sex
• level of educational attainment
• living arrangement
• health

MicMac is a generic model
12
Olivia
Household trajectory
Formal workplace trajectory
Olivia
Epros_Lux
13
State space and transitionsTransition rates
?12(t,Z)
state 1
state 2
?23(t,Z)
?13(t,Z)
state 3
?11 ?12 ?13 and ?22 ?21 ?23
14
State space and transitionsTransition rates
?12(x,t)
State 1 Healthy
State 2 Disabled
?21(x,t)
?23(x,t)
?13(x,t)
State 3 Dead
where ?11 ?12 ?13 and ?22 ?21 ?23
15
?23(x,t)
State 1 Healthy
State 2 Disabled
State 3 Reactivated
?12(x,t)
?32(x,t)
?24(x,t)
?34(x,t)
?14(x,t)
State 4 Dead
16
Pathways to first child

• States
• Transitions
• Transition rates

17
Living arrangements of women Netherlands,
Retrospective observations, OG98
18
Synthetic cohort biography State occupancies,
women, NL
19
The dynamics of cardiovascular disease Based on
the Framingham Heart Study (1948 - )
2843
1447
2382

• hCVD History of (other) CVD
• hCHD History of coronary heart disease
• hAMI history of acute myocardial infarction

A. Peeters, A.A. Mamun, F.J. Willekens and L.
Bonneux (2002) A cardiovascular life course. A
life course analysis of the original Framingham
Heart Study cohort. European Heart Journal, 23,
pp. 458- 466
20
The effect of covariates or treatment is
incorporated in the model via the transition
intensity (transition rate)
COX
baseline transition intensity
?s represent influence of covariates or
treatment on transitions between the states
21
Survival with and without cardiovascular disease
Males
hOCVD
hCHD
No hCVD

• hCVD History of (other) CVD
• hCHD History of coronary heart disease
• hAMI history of acute myocardial infarction

22
State space and transitionsWork Package 5 (D22)
Table 1. Marital status. State space and
transitions
From \ to Never married First marriage Second marriage Divorced Widowed
Never married - TR1
First marriage - TR2 TR3
Second marriage -
Divorced TR4 -
Widowed TR5 -
23
State space and transitionsWork Package 5 (D22)
Table 2. Living arrangement. State space and
transitions
From \ to at parental home Alone/with others (never in union) First union Separated (after 1st union disruption) Second union
at parental home (never in union) - TR7 TR6
Alone/with others - TR8
First union - TR9
Separated (after 1st union disruption) - TR10
Second union -
24
State space and transitionsWork Package 5 (D22)
Table 3. Fertility (own children ever born).
State space and transitions
From \ to childless 1 child 2 children 3 children 4 children
Childless - TR11
1 child - TR12
2 children - TR13
3 children - TR14
4 children -
25
State space and transitionsWork Package 5 (D22)

• Covariates
• Sex
• Men
• Women
• Education
• 1. Primary (ISCED0 pre-primary education and
  ISCED1 first stage of basic education)
• 2. Lower secondary (ISCED2 second stage of basic
  education)
• 3. Upper secondary (ISCED3 upper secondary
  education and ISCED4 post secondary non-tertiary
  education)
• 4. Tertiary (ISCED5 first stage of tertiary
  education and ISCED6 second stage of tertiary
  education)

26
Allowed covariates for each transition
TRANSITION Allowed covariates
TR1 never-married ? married (1st marriage) EDU, LIV, CHI
TR2 married (1st marriage)? divorced EDU,CHI
TR3 married (1st marriage)? widowed EDU,CHI
TR4 divorced? married (2nd marriage) EDU, CHI
TR5 widowed? married (2nd marriage) EDU, CHI
TR6 at parental home (never in union) ? first union EDU, CHI
TR7 at parental home? alone/with others (never in union) EDU, CHI
TR8 alone/ with others (never in union) ? first union EDU, CHI
TR9 first union? separated (after 1st union disruption) EDU, MAR, CHI,
TR10 alone or with other persons (after the 1st union disruption)? with a partner (2nd union) EDU, MAR,CHI
TR11 childless ? child EDU, MAR, LIV
TR12 1 child ?2 children EDU, MAR, LIV
TR13 2 children ? 3 children EDU, MAR, LIV
TR14 3 children ? 4 children EDU, MAR, LIV
Own children ever born is always coded in
only two categories childless/with children.
27
State space and transitionsWork Package 5
(D22)Episodes and dates required for each
transition
TRANSITION Episode starts at Events that cause transitions Events that cause censoring Dates required(1)
TR1 never-married ? married (1st marriage) respondents birth 1st marriage interview (ymarr,mmarr)
TR2 married (1st marriage)? divorced 1st marriage divorce death of spouse, interview (ymarr,mmarr) (ydiv,mdiv) (yved, mved)
TR3 married (1st marriage)? widowed 1st marriage death of spouse divorce, interview (ymarr,mmarr) (ydiv,mdiv) (yved, mved)
TR4 divorced? married (2nd marriage) divorce 2nd marriage death of spouse, interview (ymarr,mmarr) (ydiv,mdiv) (yved,mved) (ymarr2,mmarr2)
TR5 widowed? married (2nd marriage) death of spouse 2nd marriage interview (ymarr,mmarr) (ydiv,mdiv) (yved,mved) (ymarr2,mmarr2)
TR6 at parental home (never in union) ? first union date of birth exit from parental home for union exit from parental home for other reasons ,interview (ypartn,mpartn ()yexit,mexit)
TR7 at parental home? alone/with others (never in union) date of birth exit from parental home for other reasons exit from parental home for union, interview (ypartn,mpartn) (yexit,mexit)
TR11 childless ? 1 child respondents birth 1st childs birth interview (ych1,mch1)
TR12 1 child ? 2 children 1st childs birth 9 months 2st childs birth interview (ych2,mch2) (ych1,mch1)
28
State space and transitionsWork Package 5 (D22)
Age-specific transition rates are estimated using
Generalized Additive Models (GAM) Hastie and
Tibshirani (1990) http//en.wikipedia.org/wiki/Gen
eralized_additive_model http//www.statsoft.com/te
xtbook/stgam.html
Purpose of generalized additive models maximize
the quality of prediction of a dependent variable
Y from various distributions of the predictor
variables. Predictor variables are "connected" to
the dependent variable via a link function. GAMs
combine GLMs and linear models
Effect of covariates for each age interval
delimited by 2 knots
Cubic spline
29
Proportional effects of education on the
transition TR1, Italy
Baseline grand mean for whole same (deviation
coding) report p. 24
30
Proportional effects of education on the
transition TR1, Italy Smoothed curves
31
Age-specific rates of transition TR1, Italy
(smooth)
32
Age-specific rates of transition TR2, Italy
(smooth)
33
Age-specific rates of transition TR2, Italy
(smooth)
34
Age-specific rates of transition TR11, Italy
(smooth)
35
Transitions that can be analyzed with FFS-NL
TR1 never-married ? married (1st marriage)
TR2 married (1st marriage)? divorced
TR3 married (1st marriage)? widowed
TR4 divorced? married (2nd marriage)
TR5 widowed? married (2nd marriage)
TR6 at parental home (never in union) ? first union
TR7 at parental home? alone/with others (never in union)
TR8 alone/ with others (never in union) ? first union
TR9 first union? separated (after 1st union disruption)
TR10 alone or with other persons (after the 1st union disruption)? with a partner (2nd union)
TR11 childless ? child (only women)
TR12 1 child ?2 children (only women)
TR13 2 children ? 3 children (only women)
TR14 3 children ? 4 children (only women)
36
Age-specific rates of transition TR1, NL (smooth)
37
(No Transcript)
38
State space, several domains of life
M males
F females
   
nS never smoker
dS daily smoker
pS past daily smoker
   
I02 low level education
I34 middle level education
I56 high level education
   
nD non disabled
D disabled
39
(No Transcript)
40
TOPALSA TOol for Projecting Age profiles using
Linear Splines Joop de BeerNicole van der
Gaag(NIDI)
TOPALS is a relationale method describes
deviations from a standard schedule by linear
splines
41
Age specific fertility, 2005
Italy and average of Europe
TFR (Europe2005) 1.46 TFR (IT2005) 1.32
42
TOPALS relational model

• Assume a standard age schedule
• European average / Model schedule (Hadwiger)
• Model deviations using relative risks (RR)
• RRs for limited number of knots
• RR is average value for age interval
• Describe age pattern of RRs by linear splines
• A piecewise linear curve
• Calculate transition rates
• Multiply standard age schedule by RRs

43
Age groups and relative risks
Relative risks Relative risks Relative risks
Age IT2005 vs Europe 2005 Knots
16-21 0.48 19
22-26 0.65 24
27-29 0.78 28
30-32 0.96 31
33-40 1.90 36
41 1.50 44
is the rate at age x according to the standard
age schedule
transition rate at age x in country i
44
Linear spline through relative risks
45
Age specific fertility, 2005
TOPALS fit
TFR (Europe2005) 1.46 TFR (IT2005) 1.32
46

• Assumptions for MicMac scenarios
• Future values of transition rates
• General procedure
• - specify model curve describing age pattern
• choose age schedule that captures general
  pattern
• - specify assumptions on future values of the
  parameters
• of the model curve
• model deviations from the general pattern
• using relative risks

47
The software
48
MicMac Processor

• Pre-processor estimates the transition rates
• Processor
• Produces population projections
• Produces cohort and individual biographies
• Sequence of states
• Sojourn times
• Postprocessor
• Processes the results
• Tabulations
• Graphics
• Analysis

49
Thank youwww.micmac-projections.orgwww.demogr.m
pg.de/go/micmac