Transcript Lot 4: Spatial Analysis of interregional migration in

Interregional Migration and Land Use Pressure
B.Eiselt, N. Giglioli, R.Peckham
?
1
Acknowledgement
Based mainly on work carried out in the project:
Lot 4: “Spatial Analysis of interregional migration in
correlation with other socio-economic statistics”
Performed by JRC for EUROSTAT
from July 1998-July 1999
by: B.Eiselt, N. Giglioli, R.Peckham, A. Saltelli, T.Sorensen
2
Outline
Interregional migration modeling:
Data and Software
Spatial Interaction models
Cluster analysis
Modeling
Results
GIS based Visualization tool
Speculation on land use pressure:
Link to urban expansion
Ideas for modeling
Index for pressure
3
Data and Software
Databases:
GISCO - admin. boundaries (NUTS1 & 2)
REGIO - socio-economic data + flow matrices
Software:
SPSS 8.0 for statistical analysis
ARC-VIEW GIS (standard in E.C.)
4
Data
Country
REGIO database
Germany
1975-1990
1991-1993
Denmark
New data
Only West Germany
1991-1993
1994
1990-1993
1994
1990-1994
1990-1994
Italy
Spain
France
1975-1994
1979-1994
1968-1989
Belgium
1975-1995
1990-1994
Finland
Netherlands
1981-1995
1972-1985
1986-1995
1983-1992
1990-1994
Portugal
1980-1995
1979-1989
Missing data
Missing data
Aggregated into 1968-1974, 19751981, and
1982-1989 ( + missing data for
1982-1989)
Change of regions 1992-1995 or
missing data 1975-1991
Missing data
1990-1994
1990
Sweden
United
Kingdom
Comments
Rounded numbers + missing data
(0.0)
NB different from the CD data
(more realistic)
1990-1994
000...???
1991-1993
1994
1995
000...???
1990-1994
NB 1994 is different from the CD
data (realistic)
5
Spatial Interaction Models
Description
Exploratory analysis
Estimation of the models
Parameters interpretation
Simulation
6
Models description
The General Spatial Interaction Model has the
form
Y   e
d ij
ij
i
j
where:
• i - parameters which characterise the propensity
of each origin to generate flows;
• j - parameters which characterise the
attractiveness of each destination;
•  is a distance deterrence effect.
7
Models description
Four types:
•Double Constrained - exploring attractive properties of
destinations and repulsive properties of origins
•Origin Constrained, and Destination Constrained
- finding explanatory variables
• Unconstrained Model
- finding explanatory variables, and simulating
8
Models description
To apply the ordinary least squares fitting
we make a Logarithmic transformation of
the model in a way that the the error is
Normal distributed
lnY  a ln  b ln   cd  
ij
i
j
ij
ij
9
Correlation analysis
Analysis of correlation (Germany example)
Variables
OUT_total
IN_total
GDP
UNEMP
OUT_total
IN_total
1
0.96
GDP
0.89
1 0.93
UNEMP
-0.67
-0.57
1 -0.57
1
10
Cluster analysis
Grouping together regions displaying similar
properties,
- based on the values of:
• total inflow divided by population,
• total outflow divided by population,
• GDP per inhabitant,
• unemployment rate ( % of total workforce).
These variables are relative and are hence not influenced
by the population size of the regions.
11
Cluster analysis
Cluster analysis
13
Age structure of flows
Italy (departures=arrivals) 1993
60000
40000
30000
Italy
20000
10000
90
6_
80
y8
6_
70
y7
6_
60
y6
6_
50
y5
6_
40
y4
6_
30
y3
6_
20
y2
6_
_1
0
y1
y0
0
y6
Persons
50000
Age Group
14
Flows by clusters
15
Models !
16
Models Estimation
- Model choice:
log Y  c   log GDP   log Unem 
ij
1
i
2
i
  log GDP   log Unem   log d  e
1
j
2
j
ij
ij
- Method: Least Square and stepwise regression method
- Indicator Goodness of Fit: R2 adjusted
17
Statistics !
Kurtosis ?
30
Poisson
distribution ?
NORMALISED ??
20
Normal
distribution ?
Assumptions ?
10
Central Limit
Theorem ?

ALL
OK !
0
5
.2
10
75
9.
25
9.
75
8.
25
8.
75
7.
25
7.
75
6.
25
6.
75
5.
25
5.
75
4.
25
4.
75
3.
ln(f low )
Skewness ?
18
Models Estimation
Model estimated for Germany 1991:
Adj -R2 = 0.74
logYij = 1.767+0.934logGDPi+0logUnpi+
+0.829logGDPj+0.739logUnpj-1.156logdij
Note: the unemployment of origin is not significant
19
Simulation ?
Model fit (1991) R2 = 74%;
Forecast (1993) R2 = 65.6%
Simulation ?
Model fit (1990) R2 = 74.6%;
Forecast (1994) R2 = 55.2%
21
Simulation ?
Model fit (1990) R2 = 78.4%;
Forecast (1994) R2 = 56.8%
22
Visualization tool
23
Visualization tool
24
Visualization tool
25
Visualization tool
26
Conclusions re migration modeling
1) Some positive results. Some hope and possibilities
for modeling.
2) Need more complete and more detailed data, especially on the flows, e.g.
- age structure,
- educational level,
- cost of living, crime rate etc.
3) Need to explore and test application to other EUCountries (e.g. DK, S, Fi, NL and UK)
27
Speculation
Can we link:
migration -> land use change
?
e.g. look for correlation between:
population and urban area
- for major cities
- using satellite data to measure changes in urban perimeter,
e.g. at 5 or 10 year intervals.
As it happens there is Project MURBANDY:
http://www.riks.nl/RiksGeo/projects/murbandy/Index.htm
28
Speculation
Then we could establish the link:
GDP
->
Migration
Driving force
->
Land use pressure
Effect
Calibrate model using: Pop. : Urban area correlation
- probably different in different countries (different
habits, housing types etc)
Improve using: - age structure of flows
- education structure of flows
29
Speculation
Ideas for index of pressure:-
Population/Urban area ?
 Pop/Urban area ? = Net Flow /Urban Area
from CORINE data (grid)
30
Simulated pressure index
for year 2000 (tentative!)
31