Spatial microsimulation techniques for constructing a ... · Spatial microsimulation • = A method...
Transcript of Spatial microsimulation techniques for constructing a ... · Spatial microsimulation • = A method...
Kazumasa HANAOKAAssociate prof., Department of Geography Ritsumeikan University
Deputy director, Institute of Disaster Mitigation for Urban Cultural
Heritages, Ritsumeikan University
People. Policy. Place Seminars
Charles Darwin University
Spatial microsimulation techniques for
constructing a spatially disaggregated
population micro dataset
TokyoKyoto
Aim and Outputs
• The aim of this research is to create spatially
disaggregated microdata (synthetic microdata) of
population in a spatial microsimulation approach.
• Such synthetic microdata allows us to understand
population characteristics at the small area level for the
whole country (Japan).
• Our general purpose microdata can be used for spatial
analysis on provision of public service, retailing, disaster
risk management, etc.
Backgrounds
• In social science, disaggregated approaches using
microdata (data of decision-making units) have gained
increasing interests since 1950s.
• Microsimulation model was proposed by an economist
(Guy Orcutt) in the late 1950s. It considers heterogeneity of
individuals by using microdata.
• The model has been applied in wide disciplines but each
discipline applied it in slightly different ways. The meaning
of “microsimulation” are thus varied by discipline nowadays.
Microsimulation means:
• In economics and demographics
• Population projection and simulation based on an individual dataset
• Applications: Population projection, tax transfer, pension
• In transportation studies
• Traffic simulation (behavior of individual vehicles)
• Applications: optimization of traffic light, congestion
• In geography/geographic information science
• Small-area population estimation method and policy applications
• Application: health mapping, demand analysis in retailing
Microsimulation in Australia
• Example: Developments by the National Centre for Social
and Economic Modelling (NATSEM)
• Non-spatial and spatial microsimulation: STINMOD+, APPSIM,
SpatialMSM
Microsimulation in Geography
• Microsimulation was introduced in geography in the 1980s.
Geographers tried to add geographical element in the
simulation.
• However, spatially disaggregated microdata sets were not
readily available (even at municipality level) for
researchers in many countries.
• Thus, methods to create such microdata were studied in
1980-90s in the UK and they are often called as “spatial
microsimulation”.
Spatial microsimulation• = A method to create spatially disaggregated microdata by combining
various data sources such as census tables and survey samples
• Two major approaches
• (1) Iterative proportional fitting: A method to estimate table cell
counts based on marginal totals of benchmark tables
• (2) Reweighting by combinatorial optimization algorithm: A
method to estimate a new combination of survey samples which
agrees to marginal totals of benchmark tables at the small area level
• Others: regression type etc.Public
survey micro
data
Synthetic
microdata with
area code
Census
tables by
area
reweighting mapping
Area A Smoker Non-smoker Marginal total
Male ? ? 58
Female ? ? 42
Marginal total 30 70 100
Illustrative explanation
of combinatorial optimization
algorithm (Simulated annealing
method)
Seed microdata
(survey samples)
ResamplingGoodness-of-fit
score
Area B
Area CArea D
Area A
Benchmark
tables
tables
tabulation
Feedback to resampling
Estimated microdata
Census
tables
Datasets• Census samples for seed microdata
• 1% anonymized samples of Japanese population
census 2000 (approx. 1 million individuals)
• Resampling by household unit (no change in
household members)
• Census tables for benchmark (marginal totals)
• Small-area statistics of population census 2010
(Approx. 200,000 neighborhood areas)
Individual level
(1) sex * age
(2) sex * nationality
(3) sex * marital status
(4) sex * type of industry
(5) sex * occupation
(6) sex * work/school place
Household level
(7) building type
(8) housing tenure
(9) floor size
(10) household type
(11) household size
Total categories: 161
Small-area
statistics
Outline of our estimation method
Example: Area i in Tokyo
Select census
samples of
Tokyo only Small-area
statistics
Benchmark
tables for Area i
Area i Area k
Area j Area 1, Hokkaido
Area n, Okinawa
Area i, Tokyo
・・・
・・・
Stack all micro
data sets
Synthetic microdata of the whole Japan
= 120 million individuals
Goodness-of-fit measures
• Absolute Error(AE)
• = |census count – estimated count| for cell i
• Total Absolute Error (TAE)
• = sum of absolute error for table j
• Squared Error(SE)
• = (census count – estimated count)2 for cell i
• Total Squared Error (TSE)
• = sum of squared error for table j
For sample replacement only
Distribution of absolute error
• The 77% of all cells in estimated tables agreed with those
of benchmark tables. The 96% are within ±1.
60.0
70.0
80.0
90.0
100.0
0
5,000,000
10,000,000
15,000,000
20,000,000
25,000,000
30,000,000
0 1 2 3 4 5 6 7 8 9 10111213141516171819
20+
Freuquency
Cum ulative %
The num ber of tab le cells %
77.3%
96.8%
N= 32,584,307 tab le cells
Absolute Error = |Oi - Ei|
Household size1
person2 3 ・・・ 7+
Benchmark table 6 10 12 1Estimated table 5 10 12 0Absolute error 1 0 0 1
Cell counts by attribute
Goodness-of-fit measure• Average scores of TAE among all neighborhood areas
0 2 4 6 8 10 12 14
1
2
3
4
5
6
7
8
9
10
11
Average score of Total Absolute Error
Individual level
household level
Bad fitGood fit
Mapping results
• Average age of household head at
the neighborhood level
Age of
household
head
<=45
70+
Tabulations by any census variables
are virtually possible
Tokyo
CBD
Application
• In March 2011, the Great East Japan Earthquake gave
divesting impacts on millions of people by it huge tsunami
and radioactive contamination.
• In recent, natural disasters such as earthquake, typhoon,
heavy snow, tornado, landslide, and volcano eruption occur
almost every year in Japan.
Disaster impact assessment
• Overlay analysis of population distribution and natural
hazard distribution to present vulnerable areas.
× = vulnerability
Example1
Tohoku
Example1
Mie
Example2
Tokyo
0
0
0
0
0
7
7
7
8
0
5
9
4
6
68
5
5
7
9
4
8
5
6
4
8
7
9
6
7
8
6
89
8
9
10
1
8
4
7
8
7
9
3
4
3
6
5
8
7
8
2
4
5
8
9
6
6
9
6
8
1
0
0
3 1
3
15
14
20
42
69
85
96
60
39
61
22
13
21
27
27
21
98
28
13
28
48
46
15
41
68
69
32
26
25
22
98
15
30
36
19
21
10
16
13
10
13
12
24
21
23
90
86
29
29
18
13
11
27
13
22
41
20
11 15
11
24
29
26
36
41
15
18
2917
75
31
70
13
10
63
25
58
54
20
63
3017
73
22
58
15
20
15
26
4044
31
25
23
41
39
28
51
46
73
44
56
26
18
10
46
16
14
34
2115
1725
10
10
3125
1213
2810
2218 37
35
11
22
24
19
14102029
15
13
13
23
1611
41
2026
222127
161110
13
1510
26
4425
26
1615
1922
1327
11
34
26
27
19
25
14
20
68
21
27
23
1210
27
58
4338
20
22
93
24
32
24
12
30
10
23
59
53
34
45
15
2114
28
2525
151711
11
13
54
79
234116
22
62
63
21
44
54
7781
36
83
58
1427
7576
31
50
38
20
64
12
19
13
15
27
64
21
50
28
12
33
16
33
58
72
19
53
20
21
17
14
17
25
87
15 35
56 13
72
50 42
98
41 2236
35
34
5058
2254 99
21
58
902631
4878
7963
50
9059
3575
532869
82
10
90
346728
93
2582
58
76
40
736659
4587
6290
67
70
26
1811
55
54
33
97
51
21
43
26
2764
7298
91
1088
89
525672
271
108
125
197
179
217
243
155
117
236
355
116
191
178
260
173
172
110
171
128
109
292
701
340
476
351
786
558
336
297101
210
421
842111
239
392
156
158 152
133
105
132
154
268
104
188
508 157
129
181
125
355
117
123
南伊勢町
大紀町
志摩市
度会町
伊勢市
多気町
大台町
玉城町松阪市
明和町
鳥羽市
一関市
気仙沼市
大船渡市
陸前高田市
藤沢町
奥州市
住田町
登米市
平泉町
0
0
0
0
0
9
7
35
9
9
4
69
9
6
4
7
1
2
783
7
6 0 9
8
7
0
0
0
0
0
0
0
0
0
0
0
0
9
9
8
8
9
9
5
9
7
9
8
6
4
8
8
88
6
8
6
99
9
6
7
4
8 9
49
9
5
7
7
8
6
8
6
7
1
6
9
8
7455
55
6
6
3
7
6
7
7
9
0
6
20
42
15
19
33
38 37
302740
26
2021
2519
22
28 11
18
20
17
34
20
59171217
64
15
21 28
22 22
17
1115
1216
13
18
16
17
2722
1610
1245
19
2016 14
2328
13
11
27
19
20
29
23
11
19
16
21
3134
6514
44 53
53
28
64
2626
75
4643
3748
38 86 20
2749
2122
24
25
29
29
2735
14
12
11
11
13
54
1113
22
17
10
34
56
35
10
57
3560
5927
5458 72 16
3024 42
7141
15
15
4632 60
3637
4724
4425
24
2426
1717
21
32
23
61
47
54
20
31
17
2942
40
16
24
4033
2037
7238
33
26
24
4731
56
28
1714
1932
484647 29
142123
26
19
341515
2625
2156
545647
4244
57 4814
2251
48
85
67
525162
10
12
18
25
2220
2019
12
19
13
15
1511
10
12
13
1012
22
20
13
1819 18
2221
21 27
23
26
16
30
26
20
22
3243
2128
32
243212
1337
376026
39
24
14
21
2533
14
15
15
10 35 45
21
23
2423
20
13
1739
23
1121
14
33
13
31
25
17
1529
256231
57
42
36
38
29
52
70
342439
5753
44
25
19
19
20
1445
25
31
2629 3026
21
20
19
18
15
23
23
28
15
20
31
31 10
26
18
17
18
13
27
19
17
15
23
24
22
12
37
14
18
2024
19
31
36
36
31
59
40
17
16
18
31
25
23
1214
38
25
21
19
12
33
17
13
17
19
10
11
15
20
31
14
30
56
2442
42
17
33
23
30
49
24
38
23
20
26
34
2923
28
15
33
34
45
24
50
65
65
82
94
4695
9843
23
21
27
22
11
14
14
2816 20
18
12
12
12
21
16
14
11
11
17
31
2716
29
27
28
22
19
28
25
26
28
26
19
19
19
18
24 1520
46
16
3024
3219
18
1115
13
26
12
1444
11
11
1013
1929
332728
1526
5518
2828
1828
24
12
12
19
35
20
17
18
23
11
21
13
13
10
1515
23
20
1116
16
10
10
15
40
11
16
24
16
14
10
10
22
19
7219
16
282014
22
17
41
534224
29
10
20 28
17
13
18
3319
23
3523
12 1315
48
2135
1850
1710
46
3432
45
46
1114
28
25
18
23
26
4221
62
14
60
28
41
20
17411514
13
14
16 2871
3838
45
29
32
11
11 28
21
23
30
663246
6045
6946
20
90
92
14
2147
62
59
33
372750
1036
53 5924
4818
40
266
131
273
148
125
174
116
103
295
131
145
N
0 10km
N
Example 1 Tusnami
世帯割合(%)
0 - 10
11 - 20
21 - 30
31 - 40
41 - 50
51 - 100
浸水深(m)
0.0 - 2.0
2.1 - 5.0
5.1 - 10.0
10.1 - 15.0
15.1 -
Both size of circle and value represent
the number of households living in a
neighborhood areaNon-three generation households living in owned house
and the household head aged 65 years old and over
Proportions(%) Tsunami depth(m)
Tohoku region Mie region