Transcript Modelling propensity to move after job change using event history

Modelling propensity to move house
after job change using event history
analysis and GIS
Marie-Hélène Vandersmissen (CRAD, Laval University),
Anne-Marie Séguin (INRS-UCS), Marius Thériault (CRAD,
Laval University) and Christophe Claramunt (Naval
Academy Research Institute, France)
2nd MCRI/GEOIDE PROCESSUS Colloquium on the
Foundations of Integrated Land-Use and Transportation Models
Toronto, June 12-15 2005
Introduction
 Transportation land-use modelling must consider
decision-making behaviour of urban actors using
disaggregate data in order to relate
– Activity location, home choice, commuting and travel decision
– Household, individual and professional profiles of persons
 Probabilistic discrete choice theory is becoming the
central issue of urban and transport modelling research
– Implemented using logistic and Cox regression techniques
– Aimed at modelling individual’s and household’s behaviour
 Need for spatio-temporal GIS for analysing urban and
transport systems where
– Uncertainties exist in the system (aggregation is not straightforward)
• Emergent behaviour occurs
– Decision rules for individuals and households are intricate
– System processes are time-path and location dependent
• Future system state depends partly on past and current states
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Our project in the MCRI programme
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Purpose
 Emergent residential behaviours of individual
actors in context of profound social changes
in the work sphere
 Long term-view in the analysis of the
relationship between social changes in the
work sphere and these behaviours
Social
changes
Long-term dynamics
of residential location
behaviour
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Travel
behaviour
Objective and Research Issues
 Estimate the propensity for professional
workers to move house after a change of
workplace
– How many will move house during the
following job episode?
– For how long will they delay that decision?
– What are the factors significantly influencing
that move house decision?
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Data: The 1996 Retrospective Survey for
Quebec City
 Survey collecting, in one interview, information about all
changes occurred over a long period of time, since their
departure of the respondent’s parental home
 Spatially stratified sample of two cohorts of professional
workers
– 418 respondents living in Quebec CMA in 1995
– Two cohorts (mid-thirty and mid-forty)
• 224 women; 194 men
• 112 women and 100 men in their mid-thirty
• 112 women and 94 men in their mid-forty
– Reporting on significant events occurred during their life
time describing
• Residential trajectory (every home occupied with their location)
• Household trajectory (each change in the household’s composition)
• Professional trajectory (each change in employer, each workplace)
– Collecting dates of every change
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Complex mix of
Personal
Biography
real world
phenomena
Complex Evolution Processes
Marital
status
HOUSEHOLD TRAJECTORY
SINGLE
SINGLE
IN COUPLE
MARRIED
Leading to
SPOUSE
SPOUSE
at least one episode
UNCLE
SON
Room
Room
Change in
mate
mate
status
3
7
9
10Combining
12
21
facts17
1
4
6
describing a specific
RESIDENTIALTRAJECTORY
aspect of life
STUDIO FLAT
APARTMENT
2
Family
MOTHER
Others
persons
23 24
26
Main
home
APARTMENT
STUDIO FLAT
ROOM
ROOM
DIVORCED
TOWN HOUSE
Secondary
house
Set of relatedCHALET
lifelines
using application-specific14
semantic relationships
16
22
CAREER TRAJECTORY
STUDENT
CONSULTANT
UNEMPLOYED
PROFESSIONAL
UNEMPLOYED
Occupation
TECHNICIAN
TECHNICIAN
CONSULTANT
19
3
5
Leaving Parent's Home
8
11
13
MOTHER
18
20
25
Survey date
Time Line
Event
Episode
3
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
CONSULTANT
15
Lifeline
Location
Management of Evolution in Trajectories
 We developped a generic spatio-temporal data model to handle
historical orderings and querying patterns of facts in order to produce
flat files needed for event-history analysis
Generic part of the ST data model
HistoricalOrdering
PK
TrajectId
PK
U2,U1
TrajectName
HistoryId
Application
semantics
TrajectoryStates
Trajectories
defines
PK
LifeDimId
FK1,I1
FK2,I2
LifeStateId
TrajectId
FK1,I1 FBeforeId
FK2,I2 FAfterId
Historical ordering of
facts
uses
is after
Facts
PK
is before
FactOwners
PK
OwnerId
I1
I1
BirthDate
Survey Time
belongs to
FK2
FK3,I2
FK1,I1
I3
I4
I5
LifeStates
FactId
OwnerId
SpatialId
LifeStateId
PeriodBeg
PeriodEnd
ObsTime
belongs to
PK
LifeStateId
U2,U1
U2,U1
LifeStateName
Episode
Spatial
Individuals
Respondents
PK,FK1,FK2,U1 RespondId
I1
I1
Gender
Cohort
PK
PersonId
I1
I1
I1
Name
SurName
Gender
is
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
ActingIndividuals
is
PK
ActingId
FK2,I2
FK1,I1
PersonId
FactId
PK
SpatialId
I1
I2
Longitude
Latitude
MapInfo_MapCatalog
I1
U1
I1
I1
I1
I1
SpatialType
TableName
CoordinateSystem
Symbol
XColumnName
YColumnName
Link to Spatialware
is located at
is involved in
is
Facts : events
and episodes
Location of
facts
Spatio-temporal Query of Patterns of Facts within Trajectories
 We developped a
query interface
combining
georelational GIS
capabilities and
temporal historical
ordering of facts
using ODBC links
Specifying
spatial
distance
condition
Specifying
target
fact
Specifying
duration
condition
Specifying
time
ordering
Specifying
temporal
conditions
Specifying
spatial
location
patterns
of
factscondition
Specifying other status condition
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Methodology: Event History Analysis
 Ordinary multiple regression is ill-suited to the
analysis of biographies
– Censoring: refers to the fact that the value of a
variable may be unknown at the time of survey
– Considering time varying explanatory factors
• Need to consider time-varying information to study the effect
of job change on house moving
 Event history analysis can handle such a
problem (survival tables and logistic regression)
– The query interface enhance data restructuring
needed for this kind of statistical analysis
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Event History Analysis (Cox Regression)
 Survival tables are using conditional probabilities to
estimate the mean proportion of people experiencing some
change in their life after a significant event occurs,
computing the time delay after a specified enabling event
 Specific conditions may influence propensity to change
 Requires a combination of survival tables and logistic
regression to estimate the marginal effect of other personal
attributes on the probability that an event occurs
 Event History Analysis  to model specific variations of
the probability of state transition through time for
individuals considering independent variables describing
their personal situation on other lifelines
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Probability to move home after a job change:
probabilit y(MoveHome)  1  survival( NotMoveHome)* propensity (MoveHome)
propensity ( MoveHome)  odds( MoveHome)

odds(MoveHome)  e
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
1  odds(MoveHome)
b1 gender b2 age b3cohort bn X
e
e
e

Basic statistics
 380 respondents (on 418) had a change of job or workplace at
least once during during their career
 411 respondents moved their home at least once after
departure from parent’s home
 1056 changes of job or workplace within or towards the
Quebec CMA (321 persons)
– 458 of those changes of workplace were followed by at least
one move house during the subsequent employment episodestability of job and workplace
– 598 of those changes of workplace were not followed by any
move house during the subsequent employment episode (231
persons)
Number of pair of events (change of job-workplace versus moving house or not)
Cohort
Gender
Moving House
Not Moving House
Mid-Thirty
Men
Mid-Forty
Women
122
129
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Men
117
170
97
136
Women
122
163
Basic variables for the Event History analysis
Gender
Cohort
ChWPL_Type
ChWPL_Order
1 (Male); 2 (Female)
1 (Mid-Thirty); 2 (Mid-Forty)
1 (New Job); 2 (Change of Workplace keeping the same job)
Ordering of this change of work place among those of the same respondent (E.g. 2
means that it is the second change of workplace for this respondent)
ChWPL_Age
Age of the respondent when the change of workplace was occurring (Years)
ChWPL_Marital
Marital status when changing of workplace (1: Single; 2: Couple – marriage or free
union; 3: Separated, divorced or widow)
ChWPL_Persons Total number of persons living in the household when changing of workplace
ChWPL_Children Number of children living at home when changing of workplace
Move_House
The respondent was effectively moving house after the change of work place (1: Yes;
0: No) --- CENSORING VARIABLE
Elapsed_Time
Time elapsed between change of work place and moving house (Weeks) ---DEPENDENT VARIABLE – time elapsed at the end of the new job episode if not
moving house
MoveH_Marital
Marital status when moving home (1: Single; 2: Couple – marriage or free union; 3:
Separated, divorced or widow) – at end of new job episode if not moving home during
that period
MoveH_Persons Total number of persons living in the household when moving home – at end of new
job episode if not moving home during that period
MoveH_Children Number of children living at home when moving home – at end of new job episode if
not moving home during that period
ChWPL_Neig
Location of the new work place (1: city core; 2 old suburbs; 3: new suburbs; 4: urban
fringe)
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
PJob_Neig
Location of the previous job (1: city core; 2 old suburbs; 3: new suburbs; 4: urban
fringe; 5: outside the Quebec CMA)
PJob_Durat
Duration of the previous job episode (Years)
PJob_Regime
Employment regime at previous job location (1: Full time, >30 hours per week; 0: Part
time)
PJob_Stability
Perceived stability of employment at previous job location (1: Very stable; 2: Mostly
stable; 3: Mostly unstable; 4: Precarious)
NJob_Regime
Employment regime at new job location (1: Full time, >30 hours per week; 0: Part time)
NJob_Stability
Perceived stability of employment at new job location (1: Very stable; 2: Mostly stable;
3: Mostly unstable; 4: Precarious)
PHome_Tenure Tenure of previous home (1: owner; 2: tenant; 3: co-tenant)
NHome_Tenure Tenure of new home if any; otherwise previous tenure (1: owner; 2: tenant; 3: cotenant)
PHome_Durat
Duration of previous residential episode (Years)
PHome_Neig
Location of the previous home (1: city core; 2 old suburbs; 3: new suburbs; 4: urban
fringe; 5: outside the Quebec CMA)
NHome_Neig
Location of the new home if any; otherwise location of the old one (1: city core; 2 old
suburbs; 3: new suburbs; 4: urban fringe)
MoveH_Dist
Euclidean Distance between the old and the new residential locations (Km)
PHomeNJob_Dist Euclidean Distance between the previous home and the new job locations (Km)
NHomeNJob_Dist Euclidean Distance between the new home (if any; otherwise previous home) and the
new job locations (Km)
PJobNJob_Dist
Euclidean Distance between the old and the new job locations (Km)
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Descriptive Statistics
Change of workplace order in the respondent career
Change
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Total
Frequency
183
197
192
144
105
83
55
36
25
16
9
6
3
1
1
1056
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
%
Cumulative %
17,3
17,3
18,7
36,0
18,2
54,2
13,6
67,8
9,9
77,7
7,9
85,6
5,2
90,8
3,4
94,2
2,4
96,6
1,5
98,1
0,9
99,0
0,6
99,5
0,3
99,8
0,1
99,9
0,1
100
100
Location of the previous job episode (PJE) workplace
Frequency
City Core
316
Old Suburbs
386
New Suburbs
39
Urban Fringe
9
Outside the Quebec CMA
306
Total
1056
%
Cumulative %
29,9
29,9
36,6
66,5
3,7
70,2
0,9
71,0
29,0
100
100
Location of the new workplace (NWP)
Frequency
City Core
449
Old Suburbs
515
New Suburbs
76
Urban Fringe
16
Total
1056
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
% Cum. %
42,5
42,5
48,8
91,3
7,2
98,5
1,5
100
100,0
Change of home Neighbourhood
From or To Outside of the Quebec CMA
Core
Core --> Suburbs
Suburbs --> Core
Old Suburbs
Old Suburbs --> New Suburbs
New Suburbs --> Old Suburbs
New Subs + Fringe
Total
Frequency
100
223
41
29
484
26
14
139
1056
% Cum. %
9,47
9,4697
21,12 30,5871
3,883 34,4697
2,746 37,2159
45,83 83,0492
2,462 85,5114
1,326 86,8371
13,16
100
100
Change of tenure
Frequency
Owner --> Owner
431
Owner --> Tenant
31
Tenant --> Owner
125
Tenant --> Tenant
396
Co-tenant --> Tenant
12
Being Co-tenant
61
Total
1056
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
%
Cum. %
40,8
40,8
2,9
43,8
11,8
55,6
37,5
93,1
1,1
94,2
5,8
100
100,0
Empirical Results: 1. Cross-tables
100
90
80
X2: 1,281
ddl:1
P: 0,258
70
60
100
50
90
%
40
X2: 0,495
ddl:1
P: 0,482
80
30
70
Move home
20
60
10
1
0
0
50
Woman
Gender
40
%
Man
30
Move home
20
1
10
0
0
Mid-thirty
Cohort
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Mid_Forty
100
90
80
70
X2: 19,192
ddl:2
P< 0,000
C= 0,134
60
50
40
100
30
Move home
90
20
1
10
80
0
0
Single
X2: 89,601
ddl:4
P< 0,000
C= 0,280
70
Separated, divorced,
Married or free unio
60
Marital Status when CWP occurs
50
%
40
30
Move home
20
10
1
0
0
0
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
1
2
3
4
Number of children living at home when CWP occ
100
X2=152,63
ddl: 2
P< 0,000
C= 0,355
90
80
70
60
50
40
100
30
X2: 131,327
ddl: 4
P< 0,000
C= 0,333
90
Move home
20
10
1
0
0
Tenant
70
60
50
Co-tenant
%
Ow ner
80
40
Tenure during the Previous Home Episode (PHE)
Move home
30
20
O
0
S
e
th
ld
ity
C
c
be
ue
e
or
Q
ge
in
Fr
an
rb
s
U
rb
bu
Su
ew
N
bs
ur
ub
O
C
0
1
de
si
ut
10
C
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Type of Neighbourhood during PHE
Empirical Results: 2. Event History Analysis
Variables
Gender (0=Woman;1=Man)
Change of Home Neighbourhood
(0= From or To Ouside of the Qc CMA)
1= Core
2= Core  Suburbs
3= Suburbs  Core
4= Old Suburbs
5= Old Suburbs  New Suburbs
6= New Suburbs  Old Suburbs
7= New Suburbs + Fringe
Age
Previous Job Duration
Tenure
(0= Owner  Owner)
1= Owner  Tenant
2= Tenant  Owner
3= Tenant  Tenant
4= Co-tenant  Tenant
5= Staying Co-tenant
Previous Home Duration
Number of Children at home when CWP
Distance
New Home®New Job/Previous Home®New Job
Employment Regime at New Job Location
(0= Part Time; 1= Full Time)
Perceived Stability of Employment at New Job
(0= Very Stable)
1= Mostly Stable
2= Mostly Unstable
3= Precarious
B
SE
Wald
Sig
Exp(B)
0,307
0,102
9,105
0,003
1,359 +
-0,955
-0,163
-0,349
-0,855
-0,391
-0,247
-0,785
-0,005
0,015
0,175
0,204
0,225
0,155
0,259
0,329
0,240
0,011
0,018
49,077
29,844
0,641
2,409
30,418
2,279
0,563
10,673
0,221
0,643
0,000
0,000
0,423
0,121
0,000
0,131
0,453
0,001
0,638
0,422
0,385 0,849
0,705
0,425 0,676
0,781
0,456 0,995
1,015
1,004
0,721
0,612
1,096
0,467
-0,363
-0,204
0,261
0,186
0,175
0,346
0,272
0,025
0,098
14,799
15,107
12,221
10,063
2,940
213,393
4,327
0,000
0,000
0,000
0,000
0,002
0,086
0,000
0,038
2,728
2,057
1,845
2,993
1,595
0,695
0,815
0,052
0,019
7,284
0,007
1,053 +
0,016
0,154
0,010
0,920
1,016
0,071
0,138
0,058
0,114
0,156
0,223
0,894
0,386
0,784
0,067
0,827
0,535
0,376
0,796
1,073
1,148
1,059
+
+
+
+
-
Tests of
Model coeff.
X2: 845,29
Df: 22
Sig.: 0,000
For how long will they delay that decision?
One Minus Survival Function
at mean of covariates
1,2
1,0
,8
,6
,4
,2
0,0
-,2
-200
0
200
400
600
800
1000
1200
Elapsed time between CWP and MH (Weeks) - censoring at end of
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
One Minus Survival Function
for patterns 1 - 2
1,2
1,0
,8
,6
,4
,2
Gender
0,0
Woman
-,2
Man
-200
0
200
400
600
800
1000 1200
Elapsed time between CWP and M H (Weeks) - censoring at en
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
Discussion and Conclusion
 Results given by Event History Analysis:
– How many will move house during the following job episode?
• On 418 respondents, 271 moved home after a job change (64,8%)
– For how long will they delay that decision?
• Probability of changing home after a job change =0,2 after ~2 years
– What are the factors significantly influencing that move house decision?
• Tenure
–
–
–
–
•
•
•
•
•
Co-tenant  Tenant
Owner  Tenant
Tenant  Owner
Tenant  Tenant
Gender (man)
Increased Distance home  job
Number of Children
Previous home duration
Change of Home Neighbourood
– New Suburbs + Fringe
– Old Suburbs
– Core
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005
+
+
+
+
+
+
-
 Retrospective Survey
– Inaccuracy of responses (limitations of human memory
with elapsed time)
– Memory distorsions (individual’s account of the event)
– But people tend to remember major events (year of
residential move, job change)
– Results reflect situation in 80’s and 90’s
 To the best of our knowledge, this type of
application is original (residential move after a job
change
– Positive contribution to transportation land-use
modelling (Quebec)
– The query interface could be also used to analyse
patterns of activity/travel decision coming from our
panel surveys (Quebec & Toronto) and OD surveys
– Next stage: Elaborate separate models for owners and
tenants
2nd MCRI/PROCESSUS Colloquium, Toronto, June 12-15, 2005