Generalized gravity model for human migration
Hye Jin Park1, Woo Seong Jo2,3,4, Sang Hoon Lee5,∗, and Beom Jun Kim2,†
1Department of Evolutionary Theory, Max Planck Institute for Evolutionary Biology, 24306
Pl¨on, Germany
2 Department of Physics, Sungkyunkwan University, Suwon 16419, Korea
3 Northwestern Institute on Complex Systems (NICO), Evanston, Illinois 60208, USA
4 Kellogg School of Management, Northwestern University, Evanston, Illinois 60208, USA
5 Department of Liberal Arts, Gyeongnam National University of Science and Technology,
Jinju 52725, Korea
E-mail: ∗lshlj82@gntech.ac.kr
E-mail: †beomjun@skku.edu
Abstract.
The gravity model (GM) analogous to Newton’s law of universal gravitation has
successfully described the ﬂow between diﬀerent spatial regions, such as human migration,
traﬃc ﬂows, international economic trades, etc. This simple but powerful approach relies
only on the ‘mass’ factor represented by the scale of the regions and the ‘geometrical’ factor
represented by the geographical distance. However, when the population has a subpopulation
structure distinguished by diﬀerent attributes, the estimation of the ﬂow solely from the coarse-
grained geographical factors in the GM causes the loss of diﬀerential geographical information
for each attribute. To exploit the full information contained in the geographical information
of subpopulation structure, we generalize the GM for population ﬂow by explicitly harnessing
the subpopulation properties characterized by both attributes and geography. As a concrete
example, we examine the marriage patterns between the bride and the groom clans of Korea in
the past. By exploiting more reﬁned geographical and clan information, our generalized GM
properly describes the real data, a part of which could not be explained by the conventional
GM. Therefore, we would like to emphasize the necessity of using our generalized version of
the GM, when the information on such nongeographical subpopulation structures is available.
arXiv:1805.10422v2  [physics.soc-ph]  18 Sep 2018

Generalized gravity model for human migration
2
1. Introduction
For decades,
the gravity model (GM) has successfully explained ﬂows between
geographically separated two regions such as traﬃc ﬂow [1, 2, 3, 4, 5], international economic
trades [6, 7], and human migration [8, 9]. The GM is named after Newton’s law of universal
gravitation because of the similarity in the formula: a certain type of ﬂow between two regions
is proportional to the product of ‘mass’ of each region and inversely proportional to a certain
power of distance between the regions. We interpret the mass depending on contexts; we
can quantify the relative importance of regions in human migration from their population
sizes, and the relative importance of countries in international trades from their economic
scales. This simple but powerful model has succeeded in interpreting the real world. For
instance, the GM indeed accurately describes the empirical data of daily human mobility in
multiscale mobility networks [10]. It also nicely explains the inter- and intra-city traﬃc ﬂows
in Korea [11], along with the passenger ﬂows in the Korean subway system [12].
However, those examples only concern the spatial aspect of population. We can easily
imagine more complicated situations such as the population ﬂow between other attributes
than the spatial or regional attributes, when the population at a given region consists
of subpopulation structures. The subpopulation structures characterized by attributes for
population ﬂows can be ethnic groups, income levels, etc.
In particular, when spatial
movement of the subpopulation belonging to one attribute to another subpopulation takes
place, the ﬂow between attributes becomes relevant. This applies not only to the population
ﬂow but also to the international trade, for instance.
Goods are transferred in diﬀerent
economic sectors as the attributes, and one may aim at estimating the ﬂow between economic
sectors. One such previous attempt to apply the GM to explain ﬂow between attributes is
[13] partly by a subset of the authors of this paper. However, the results have revealed the
limitation of the conventional GM when the geographical location of population center of a
clan is not representative. The limitation stems from the process of signiﬁcant coarse graining
of the detailed geographical information of clans into a single point (the population center).
In this paper, we exploit detailed information on the population substructures by
generalizing the GM, rather than coarse graining the information. First, we formulate the
generalized gravity model (GGM) to take the full information available on the subpopulation
(characterized by both geographical information and attribute information) ﬂows. As we
show in ﬁgure 1(a), when subpopulations are distinguished not only by geographical regions
but also by attributes, ﬂows between regions or attributes can be calculated by properly
taking subpopulation ﬂows without information loss (see ﬁgures 1(b) and (c), respectively).
We also would like to emphasize the necessity of calculating subpopulation ﬂows, because
the population ﬂows calculated from the coarse grained population data and those from
individual subpopulation data are not equivalent. As a concrete example, we apply the GGM
to marriage records combined with census data. The results show that it eﬀectively captures
the geographical constraint imposed in the marriage patterns in the past, in contrast to the
GM.
The paper is organized as follows.
We ﬁrst formulate the GGM in section 2.
We

Generalized gravity model for human migration
3
(b)
A
B
i
i
j
j
i
B
i
A
B
C
j
A
B
C
j
(a)
i
j
A
B
C
D
A
(c)
+
iCM
+
jCM
D
D
i
j
i    j
i    j
+
iCM
+
jCM
GM
GGM
GM
GGM
Figure 1. (a) We present the schematic ﬁgure for two spatial distributions of attributes, i and
j, and ﬂows between subpopulations. Each attribute has its own spatial population distribution
represented by the color gradient with the center of mass, e.g., iCM or jCM (marked by the
× symbol). Since the population for a given region is again distinguished by its attribute,
each subpopulation is distinguished by both region and attribute indices denoted by uppercase
letters, A, B, C, and D, and lowercase letters, i and j, respectively. We illustrate all of the
possible ﬂow combinations between subpopulations, and we use the rectangular and oval
boundaries to distinguish regions and attributes, respectively. (b) We show the comparison
of ﬂows between regions in the viewpoints of GM and GGM. The GM described on the
left considers the ﬂow between regions. It integrates the population size at a given region
ﬁrst to calculate the ﬂow, while the GGM on the right ﬁrst considers subpopulation ﬂows
and integrates them. As we show in section 2, the results from the GM and GGM are not
equivalent. (c) We compare the ﬂows between attributes in the GM and GGM. The GM on the
left takes into account the ﬂow between attributes i and j from the centers of populations as
in [13], which causes some information loss due to the coarse graining. In contrast, we keep
the entire information available by integrating all of the subpopulation ﬂows in the GGM (this
paper).
introduce our data set in section 3 and apply the model to this data. The result in section 4
demonstrates that the GGM indeed captures the geographical information not available from
the GM. Finally, we conclude our work in section 5.
2. The GGM
Our derivation of the GGM is a natural extension of using the maximum entropy principle
to derive the GM [14, 15], where we replace regional indices with both regional and
attribute indices.
We provide our step-by-step derivation in this section partly for a
pedagogical purpose and the self-containedness of this paper, but most importantly, we
can directly demonstrate the problem of using the coarse-grained population data during
the derivation. The maximum entropy principle is the way to estimate a real probability
distribution by maximizing entropy. In particular, this method is useful for systems with
many degrees of freedom because it focuses on only a few macroscopic quantities. The real

Generalized gravity model for human migration
4
probability distribution is estimated by the maximum entropy principle from the agreement
of those observed quantities. Each Lagrange multiplier corresponding to each constraint in
maximization gives the corresponding model parameter.
Let us start from the number N(ij)
AB of people who move from region A with attribute i to
region B with attribute j, where we take the convention of uppercase letters as the subscript
for the region indices, and the lowercase letters as the superscript for the attribute indices.
The sets of attributes {i} for the sender side and { j} for the receiver side can be diﬀerent. For
example, {i} and { j} can be the education level and the income level, respectively, when we try
to describe the ﬂow of employment from one city’s education system to another city’s industry.
Our formalism adopts the discrete indices and summation, but it should be straightforward to
deal with the continuous cases by using continuous variables and integration.
The total number N(i)
A of people moving from region A with attribute i to anywhere with
any attribute is then
N(i)
A =
X
j,B
N(i j)
AB .
(1)
In the same way, the total number ˜N( j)
B
of people who arrive at region B with attribute j from
anywhere with any attribute is given by
˜N( j)
B
=
X
i,A
N(i j)
AB .
(2)
When people move, they have to pay the cost, which is naturally a function of the distance
between two regions, among other factors. We denote the cost to move from (i, A) to ( j, B)
for each unit of movement by c(ij)
AB . Then, the total moving cost C is the following weighted
sum,
C =
X
i,A,j,B
N(i j)
AB c(i j)
AB .
(3)
We can then write down the number W of all possible arrangements of travelers
considering the multiplicity factor N(ij)
AB as
W =
N!
Y
i,A,j,B
N(i j)
AB
,
(4)
where N is the total number of moving people, N = P
i,A,j,B N(i j)
AB . In the entropy maximization
scheme, {N(ij)
AB } is estimated from maximizing the Boltzmann entropy kB log W (equivalently
maximizing W) with constraints. We consider three constraints under which W is maximized:
the outﬂows {N(i)
A }, the inﬂows { ˜N( j)
B }, and the total moving cost C represented in equations (1)–
(3). For this optimization problem under given constraints, we have to use the Lagrange
multiplier method, i.e., to ﬁnd the stationary point of the Lagrangian
L({N(ij)
AB }) =
N!
Q
i,A, j,B N(ij)
AB
+
X
i,A
λi
A
N(i)
A −
X
j,B
N(i j)
AB
+
X
j,B
˜λj
B
˜N( j)
B
−
X
i,A
N(i j)
AB
+ γ
C −
X
i,A, j,B
N(i j)
AB c(i j)
AB
,
(5)
where λi
A, ˜λ j
B, and γ are the Lagrange multipliers for each constraint.
This problem is
essentially the recap of the derivation for the most probable distribution in terms of a given

Generalized gravity model for human migration
5
energy value, from the standard formalism of canonical ensemble in statistical mechanics, so
the readers may check the details in any standard statistical mechanics textbooks such as [19].
The moving cost for each movement and the γ parameter here play the roles of energy and
inverse temperature there, respectively. The solution of maximizing equation (5) is given by
N(i j)
AB ∝N(i)
A ˜N(j)
B e−γc(ij)
AB .
(6)
Note that all λi
A and ˜λ j
B become unity from the constraint itself (the mass conservation), so
there is only one free parameter γ, which is determined by the real data. Later, we will
speciﬁcally choose the γ value that minimizes the error between the model and real data.
The ﬂow from region A to region B usually decays as a function of the distance between
them due to the obviously rising cost, and thus we have to choose the cost function c(r) as an
increasing function of the distance r between two regions. Conventionally, we set the form
c(r) ∝ln r [20], which leads equation (6) to
N(i j)
AB ∝N(i)
A ˜N( j)
B
(rAB)γ .
(7)
Note that, though the numbers of leaving or arriving people, N(i)
A
and ˜N(j)
B , are not the
population size P attribute
region
at each region and attribute, generally those are assumed to be linear
to the population sizes [N(i)
A ∝P(i)
A and ˜N( j)
B
∝P(j)
B ], yielding
N(i j)
AB ∝P(i)
A P( j)
B
(rAB)γ .
(8)
In this case, we can reproduce the GM for the ﬂow between two spatial regions
NAB ≡
X
i,j
N(ij)
AB ∝
X
i
P(i)
A
X
j
P(j)
B
(rAB)γ
= PAPB
(rAB)γ ,
(9)
where PA and PB are total numbers of people who live in region A and region B, respectively.
However, when N(i)
A and ˜N( j)
B
are nonlinear with respect to the population sizes such as in [10],
i.e., N(i)
A
∝[P(i)
A ]α and ˜N( j)
B
∝[P(j)
B ]β, the population ﬂow from region A with attribute i to
region B with attribute j: N(ij)
AB ∝[P(i)
A ]α[P(j)
B ]β/(rAB)γ. In this case, the ﬂow from region A to
region B regardless of attributes: NAB cannot be explained by the GM unless α = β = 1 or the
population contains only one attribute, because
NAB ≡
X
i, j
N(ij)
AB ∝
X
i
[P(i)
A ]α X
j
[P(j)
B ]β
(rAB)γ
, [PA]α[PB]β
(rAB)γ
.
(10)
Note that the conventional GM provides the upper or lower bounds: P
i[P(i)
A ]α ≤[P]α for α > 1
and P
i[P(i)
A ]α ≥[P]α for α < 1, by using the convexity or concavity of the functional form.
In parallel, summing up N(ij)
AB for all of the regions gives the number of moving people
from attribute i to j,
N(ij) ≡
X
A,B
N(i j)
AB ∝
X
A,B
N(i)
A ˜N( j)
B
(rAB)γ .
(11)

Generalized gravity model for human migration
6
This is not reducible to the GM either, because
X
A,B
N(i)
A ˜N( j)
B
(rAB)γ ,
1
r(iCM, jCM)γ
X
A,B
N(i)
A ˜N(j)
B ,
(12)
where r(iCM, jCM) is the distance between the centers of population of i and j, as shown
in ﬁgure 1. The only case when the two expressions actually coincide is the assumption
implicitly made in [13]—the population of each attribute i is treated as the ‘point mass’ located
in a single location in space, namely, iCM. The GGM estimates the ﬂow between attributes
without such a coarse graining process involving the information loss. Hence, the GGM is the
correct way to handle subpopulation structures when it comes to the GM of population ﬂow.
We later show that it indeed eﬀectively captures the geographical constraints for the marriage
ﬂow in the past obtained from the data, which was not possible with the coarse-grained version
of the GM due to the widely distributed population of the clans [13].
3. Data sets
In the traditionally patriarchal culture of Korea after around 17th century, a bride usually
moved to her groom’s place in the past, once they got married. We treat this type of migration
caused by the marriage as our main data of human migration. By applying the GGM to
marriage patterns between clans in the past, we estimate the geographical constraints. We
take the real marriage ﬂow O(ij) from the bride clan i to the groom clan j from the family book
data called jokbo. We present more details in section 3.1. To compute the model ﬂow, we
extract the distance rAB between two regions and the distribution of the population for each
clan, N(i)
A
(the bride side) and ˜N( j)
B
(the groom side), from the modern census data in 1985,
2000, and 2015—the three particular years when the information on the regional distribution
of each clan’s population is available.
We measure the distance rAB based on geographical coordinates of the regions using the
Google maps application programming interface [21]. The traveling distance within the same
region rAA is estimated as the square root of the region’s area. We assume that the size of
the moving populations, represented by N(i)
A and ˜N(j)
B , are proportional to that of the resident
populations (from the census data) of the corresponding clans living in the corresponding
regions, so we just take the face values of populations in the census data and treat them as
the migrating population for simplicity. As argued in [13], we use this modern population
data to estimate the past migration ﬂow between clans in jokbo data, based on the fact that the
proportion of each clan living in each region with respect to the total population of Korea has
been relatively steady.
3.1. Jokbo data
Jokbo, or the Korean family book, records the members of paternal lineage and each
member’s spouse and children. Even though a bride does not change her family name after
marriage in Korean culture, she was (and still is in many conservative families) considered

Generalized gravity model for human migration
7
Table 1.
The volume of the jokbo and the census data. The number of bride clans and the
total number of entries in each jokbo are counted based on the existing clans in the census data.
Note that we count all brides in the jokbo whether it includes birth and death dates or not, and
thus the volume can be diﬀerent from the previous research [13, 16, 17, 18]. In addition, the
population sizes of the groom clans in the census data are presented.
jokbo clan
jokbo data
census data (population size)
number of bride clans
number of entries
year 1985
year 2000
year 2015
1
1 755
155 392
3 892 342
4 324 478
4 456 700
2
1 077
59 588
47 383
61 650
78 607
3
1 149
54 377
200 334
232 753
298 092
4
901
25 542
25 115
25 667
34 802
5
782
39 405
231 289
238 505
324 507
6
804
25 343
21 756
21 536
27 343
7
1 723
189 158
343 700
380 530
445 946
8
607
12 846
15 539
17 939
20 484
9
356
5 146
103 220
123 688
163 610
to belong to the groom’s family after marriage. The key element of jokbo for our research
is the fact that it records information of the female spouse’s original clan including the
information on its geographical origin. Each clan has its own jokbo, which is passed down to
descendants. Previously, the distributions of clans in Korea have been studied based on ten
jokbo data [16, 17, 18]. Marriage patterns using the same data set have been studied in [13]
with the GM framework under the assumption described in the left ﬁgure of ﬁgure 1(c).
We also use the same ten jokbo data set, but at this time we merge two jokbo among
ten because those two are diﬀerent subgroups of the same clan (because our attribute unit
is the clans), which results in the total number of nine distinct jokbo used in our analysis.
We count how many brides from clan i married the grooms from clan j, the owner of jokbo,
and treat the number of brides as the real migration ﬂow O(i j) from the bride clan i to the
groom clan j. Each jokbo contains between 5 146 and 189 158 marriage entries (see table 1
for detailed statistics). We index the jokbo in the ascending order of the value of γ( j)
opt (that will
be introduced in section 4) predicted from the 2015 census data. There is a single case of a
tie, and we break it by using γ( j)
opt from the 1985 data.
3.2. Census data
We assume that the outﬂow N(i)
A from (i, A) and the inﬂow ˜N(j)
B to (j, B) are proportional to their
population sizes, P(i)
A and P( j)
B , to predict the ﬂow N(i j) from equation (11). The population
size of each clan residing in each region is taken from the Korean census data [22], where
the spatial resolution of the data is determined by the set of 194 administrative regions. In
particular, we use the census data in the years 1985 and 2000 as in [13], and the new data in
2015. We present the detailed population statistics for each groom clan in the census data in
table 1.
Due to the changes in the administrative boundaries over 30 years (between 1985 and
2015), we have generated the common set of 194 administrative regions for the three diﬀerent

Generalized gravity model for human migration
8



     









 


 


	




     















	




     















	




     















	




     















	




     















	




     















	




     















	




     













	

Figure 2. The error landscapes as the function of γ and the scatter plot of real and estimated
marriage ﬂows at γ(j)
opt for each groom clan j, with the census data in 2015. We index the groom
clans according to γ(j)
opt in the ascending order, from 1 to 9. The panels (a)–(i) correspond to
the groom clans 1–9, respectively. For the actual error landscape plots, we use the normalized
error E(j)(γ)/E(j)(γ = 0) with respect to the γ = 0 case. The vertical dashed lines in the error
landscapes indicate the γ( j)
opt value that gives the minimum value of E(j).
years, where we have uniﬁed the administrative regions whose boundaries had been changed,
following the procedure of [13] to unify the administrative regions (for the two diﬀerent years:
1985 and 2000, in [13]).
4. Results
We apply the GGM expressed in equation (11) to the marriage patterns in the past. The real
number of marriage entries {O(ij)} are listed in the jokbo data, and we compare the predicted
ﬂow {N(i j)} from the model with {O(ij)}. For each groom clan (corresponding to each jokbo
clan) j, the diﬀerence is quantiﬁed by the error
E( j) =
sX
i
E(i j)2 ≡
sX
i
O(i j) −N(i j)2 ,
(13)
calculated from the list of the bride clans {i}. Note that we discard the self migration ﬂow
E( j j) when we calculate E(j), because the marriage between the same clans was forbidden
in the past and it is indeed signiﬁcantly underrepresented as reported in [13]. In practice,
we also checked that ignoring E( jj) does not make much of a diﬀerence in our results. The
proportionality factor for equation (11) is calculated by minimizing E(j) at a given value of
γ( j). The optimal value γ( j)
opt is assigned as the γ( j) value that minimizes the error E(j). In this

Generalized gravity model for human migration
9
 0
 2
 4
 6
 8
 10
1
2
3
4
5
6
7
8
9
γopt
(j)
groom clan
1985
2000
2015
Figure 3. The estimated exponent γ(j)
opt of distance in equation (11) from the census data in
1985, 2000, and 2015, for each groom clan j. The horizontal axis indicates groom clans, and
we use three diﬀerent types of symbols to distinguish the results for each year. We shade every
other column for better readability.
Table 2. The results of γ(j)
opt for each groom clan j, from 1985, 2000, and 2015 census data,
respectively. For the comparison between the GGM and the GM, we provide the values of ∆e(j)
(%) in equation (14) as the percentage, representing the relative performance of the GGM. We
also characterize the population distribution of each clan by measuring the dispersion ∆R(j)
(km) in equation (15) and the eﬀective number of occupied regions n( j) in equation (16).
groom
clan j
year 1985
year 2000
year 2015
γ(j)
opt
∆e( j)
∆R(j)
n(j)
γ( j)
opt
∆e( j)
∆R(j)
n(j)
γ( j)
opt
∆e( j)
∆R(j)
n(j)
1
0.1
0.02
151.35
104.23
0.2
0.17
148.54
81.43
0.2
0.38
146.32
76.65
2
1.7
9.98
118.39
43.15
1.7
8.50
118.43
43.18
1.7
7.57
117.83
47.21
3
1.6
17.81
149.91
71.31
1.7
18.99
142.11
62.44
1.8
16.81
139.23
60.14
4
1.7
15.01
131.46
85.29
1.7
15.17
125.76
67.34
1.8
13.51
125.52
64.48
5
1.3
0.52
143.01
54.98
1.8
1.28
146.06
51.84
2.0
2.02
146.76
62.44
6
1.7
9.43
141.09
60.74
2.0
10.12
137.44
50.50
2.1
9.24
134.61
51.60
7
5.3
6.05
155.77
91.37
5.6
9.69
152.98
74.21
6.2
10.32
149.64
71.26
8
6.0
2.88
140.40
63.56
6.6
6.64
136.73
61.83
6.9
7.08
133.10
57.67
9
7.2
2.18
139.94
87.32
7.3
2.54
136.68
70.49
7.7
2.40
134.97
69.69
case, we vary the γ( j) value from 0 to 10 with the resolution of 0.1. The obtained γ( j)
opt value
indicates the geographical constraint for the brides’ migration to the groom clan j.
In ﬁgure 2, on the left side of each panel (corresponding to each groom clan j), we show
the error in equation (13) as a function of γ( j), via the fact that N(i j) is a function of γ( j). On
the right side of each panel, we also show scatter plots comparing the real ﬂow O(i j) versus
the predicted ﬂow N(ij) from the GGM at a given γ( j)
opt value with the guideline corresponding
to O(i j) = N(ij). The predicted ﬂow N(i j) and the number of entries O(i j) in jokbo are indeed
close to the O(ij) = N(ij) line. Since the results from 1985, 2000, and 2015 census data are
qualitatively the same, we only show the results in 2015. Except for the clan 1, the GGM
actually captures nonzero γ( j)
opt values, while the exponent γ( j)
opt always vanishes when we use
the GM for all of the jokbo [13].
We calculate N(ij) from each census data in 1985, 2000, and 2015, so we obtain the

Generalized gravity model for human migration
10
 0
 4
 8
 12
 16
 20
 1
 2
 3
 4
 5
 6
 7
 8
 9
∆e(j) (%)
groom clan
Data
Shuffled Data
Figure 4.
The performance of the GGM expressed as the normalized reduced error in
equation (14) from the census data in 2015 for each groom clan j, compared with the
model performance from geographically scrambled data. The values for the real data and the
average values from the shuﬄed data are shown as the large purple diamonds and the small
green diamonds with the error bars representing the standard deviation over 100 realizations,
respectively. As in ﬁgure 3, The horizontal axis indicates groom clans and we shade every
other column for better readability.
three γ(j)
opt values corresponding to each year, for each groom clan (see ﬁgure 3). From the
similarity of error landscapes in each census data, the γ(j)
opt values for diﬀerent years are not
much diﬀerent for a given groom clan j (check table 2 for details), which indicates that the
results of γ(j)
opt are temporally robust for a given groom clan. It is also interesting to note
that many γ(j)
opt values are around 2 corresponding to the same formula with demographic
gravitation introduced in [2]. Most importantly, compared with the GM results, i.e., γ( j)
opt ≃0
for all of the groom clans [13], the GGM indeed yields nonzero γ( j)
opt values except for the clan
1. It implies that the GGM actually captures the information of the geographical constraint on
the ﬂow, in contrast to the GM where it is hard to capture this geographical information due
to the coarse graining, i.e., treating the population of a clan as a point particle at a center of
mass. On the contrary, the GGM uses much more detailed information of the subpopulation
structure that eventually leads us to capture the actual geographical constraint imposed on the
ﬂow.
To validate our model, we measure the performance of our model compared with the GM
using the normalized reduced error with respect to the case of no geographical constraint, i.e.,
γ = 0, deﬁned as
∆e(j) =
h
E(j) (γ = 0) −E(j) 
γ = γ( j)
opt
i
E(j) (γ = 0)
.
(14)
The normalized reduced error ∆e(j) quantiﬁes the improvement of performance by using the
GGM compared with the GM, which results in γ( j)
opt ≃0 for all of the clans [13]. Large values
of ∆e( j) indicate signiﬁcance of geographical constraints in the migration ﬂow. Except for
clans such as j =1, 5, and 9, the normalized reduced error ∆e( j) ≃10%, as shown in ﬁgure 4.

Generalized gravity model for human migration
11
To demonstrate the statistical signiﬁcance of geographical information of the data, we shuﬄe
regional indices for the groom clan j to obtain the corresponding surrogate γ( j)
opt values, also
shown in ﬁgure 4.
For the exceptional cases of j = 1, 5, and 9, we suspect the lack of geographical
information in the data itself, as we argue.
To test the statistical signiﬁcance of spatial
correlation in the data, we shuﬄe the regional indices to scramble geographical information.
We examine the result of our model in this shuﬄed data, which is shown in ﬁgure 4. It
supports that the GGM extracts more geographical information than the GM, by capturing
the nonzero γ exponent. If the shuﬄed data gives similar results to those from original data,
the original data contains a small amount of geographical information. As shown in ﬁgure 4,
this situation precisely happens for the clans j = 1, 5, and 9, whose original and shuﬄed data
give similar results. In other words, the data corresponding to those three clans originally
contain less geographical information than the other clans. The small ∆e( j), therefore, does
not come from the GGM but from the clan data itself. Hence, we conclude that as long as the
data has enough geographical information, the GGM eﬀectively extracts the corresponding
information.
As mentioned above, geographical information of the population distribution is closely
related to the model performance and the statistical signiﬁcance of the nonzero γ( j)
opt values.
To quantify the geographical information in the distribution of clans more systematically, we
introduce two measures: the dispersion that quantiﬁes how strongly localized the populations
are, and the homogeneity that focuses on how uniformly the populations occupy distinct
regions. For the latter, we use the concept of the eﬀective number of occupied regions based
on the R´enyi entropy for a given probability distribution, as in [23]. We deﬁne the dispersion
∆R(j) from the centroid R( j) of the clan j by taking population fractions as weights:
∆R( j) =
sX
A
f ( j)
A

rA −R( j)

2,
(15)
where rA is the location of the administrative region A. The population fraction f (j)
A
is the
population of clan j living in A divided by the total population of clan j, and || · · · || is the
Euclidean norm. The population centroid of clan j is then R( j) = P
A f (j)
A rA. As the concept of
moment of inertia or radius of gyration [13], ∆R(j) measures how (geographically) widely
a certain clan is distributed. The eﬀective number of occupied regions is deﬁned as the
reciprocal of the heterogeneity quantiﬁed by the second moment of the population fraction,
given by
n( j) =
1
X
A
h
f ( j)
A
i2.
(16)
In contrast to ∆R(j), n( j) measures how many of distinct regions (regardless of their
geographical location) a certain clan occupies eﬀectively; there are scaling relations for
extreme cases: n(j) ≃the total number of administrative regions when the clan j is uniformly
distributed to the entire set of administrative regions, while n(j) ≃1 when the clan j is almost
exclusively living in a single particular administrative region [23].

Generalized gravity model for human migration
12
 0
 50
 100
 150
 200
∆R(j) (km)
 0
 50
 100
 150
n(j)
 0
 4
 8
 12
 16
 20
Figure 5. The density plot of dispersion ∆R( j) and the eﬀective number of occupied regions
n( j) for all of the clans in the 2015 census data. The nine orange diamonds correspond to the
groom clans in the jokbo data.
The dispersion ∆R( j) and the number n( j) of occupied regions are usually positively
correlated. However, ∆R(j) can be large even when n(j) is small, e.g., when the clan has
multiple localized residential regions.
Hence, we use both measures for more accurate
identiﬁcation of the population distributions. For all of the 788 clans in the 2015 census
data, we measure both ∆R( j) and n( j) and present them as the density plot in ﬁgure 5. Among
all of the clans listed in the census data, all of the groom clans corresponding to the jokbo data
have relevantly large values of dispersion and the eﬀective number of occupied regions. Note
that the combination of large ∆R( j) and small n(j) (as discussed before) is observed indeed,
while the combination of small ∆R( j) and large n( j) does not appear, as shown in ﬁgure 5.
This contrast hints the existence of such multiple localized residential regions, which is also
discussed in [13].
Finally, we compare the relative performance of GGM with these two measures,
presented in ﬁgure 6. There is a trend of increasing ∆e(j) when the population distribution
has more geographical information, low dispersion and low eﬀective number of occupation
regions. The Pearson correlation coeﬃcients between ∆e( j) and the dispersion R(j) or the
eﬀective number n(j) of occupied regions are −0.270 and −0.213, respectively, implying the
anticorrelation.
This observation again conﬁrms that when the data includes meaningful
geographical information, the GGM captures the geographical constraint, while the GM may
not.

Generalized gravity model for human migration
13
1985
2000
2015
 100
 120
 140
 160
∆R(j) (km)
 30
 60
 90
 120
n(j)
 0
 4
 8
 12
 16
 20
Figure 6. The same ∆R( j)–n( j) diagram as ﬁgure 5, but only with the groom clans, using
the census data in the three diﬀerent years. The color of the points represents ∆e( j), and the
diﬀerent symbols indicate the results from the census data in diﬀerent years.
5. Conclusions and discussion
In this paper, we formulate the GGM to properly take subpopulation structures for human
migration, by keeping the entire geographical distribution of the subpopulations. The key
aspect of our point is that we need to calculate individual subpopulation ﬂows, before trying
any geographical coarsening. To test the validity of the GGM, we investigate the marriage
patterns of Korea in the past. Applying our model to the marriage pattern, we identify the
geographical constraint. The results demonstrate that the GGM captures the subpopulation
aspect of the data without the information loss occurred in the GM.
We believe that our approach is applicable to a wide range of research on population
dynamics. Moreover, we would like to point out that the GGM is in fact even more general
than the treatment for our particular data set, e.g., the diﬀerent types of attributes for the
departure and arrival places by taking the diﬀerent sets {i} , { j}. For instance, the attribute in
the departure place for education can be the education level of people, while the attribute of
the arrival place for work can be the income level. Furthermore, if we release the constraints
α = 1 and β = 1, one can also allow the nonlinear mass relation. In this case, the GGM
becomes particularly important to prevent the loss of information because the scale factor, in
addition to the distance factor, also has the nonlinear relation with the ﬂow.
Finally, this scheme can be extended for multiple types of attributes that can be
represented by the attribute vectors. For instance, people living in the region A with the
attribute i = (ei, mi) representing the education level ei and the income level mi can move to
the region B with j = (e j, mj). We hope to extend this type of general scheme for a wide

Generalized gravity model for human migration
14
variety of diﬀerent data sets of human (and possibly nonhuman) migration or ﬂow patterns in
the future.
Acknowledgments
S.H.L. was supported by Gyeongnam National University of Science and Technology Grant
in 2018–2019. B.J.K. was supported by the National Research Foundation of Korea (NRF)
grant funded by the Korea government (MSIT) (No. 2017R1A2B2005957). We appreciate
the anonymous referee for pointing out that the conventional gravity model plays the role of
upper or lower bounds.
References
[1] Reilly W J 1931 The Law of Retail Gravitation (New York: Knickerbocker Press)
[2] Stewart J. Q. 1948 Sociometry 11 31–58
[3] Carrothers G A P 1956 J. Am. Inst. Plan. 22 94–102
[4] Erlander S and Stewart N F 1990 The Gravity Model in Transportation Analysis (Utrecht: Brill Academic
Publishers)
[5] Rodrigue J P, Comtois C and Slack B 2009 The Geography of Transport Systems (London, New York:
Routledge)
[6] Tinbergen J 1962 Shaping the World Economy; Suggestions for an International Economic Policy (New
York: Twentieth Century Fund)
[7] Nello S S 2009 The European Union: Economics, Policy and History (New York: McGraw-Hill)
[8] Ravenstein E G 1885 Journal of the Statistical Society of London 48 167–235
[9] Anderson J E 2011 Annu. Rev. Econ. 3 133–60
[10] Balcan D, Colizza V, Gonc¸alves B, Hu H, Ramasco J J and Vespignani A 2009 Prof. Natl. Acad. Sci. 106
21484–9
[11] Jung W S, Wang F and Stanley H E 2008 Europhys. Lett. 81 48005
[12] Goh S, Lee K, Park J S and Choi M Y 2012 Phys. Rev. E 86 026102
[13] Lee S H, Ffrancon R, Abrams D M, Kim B J and Porter M A 2014 Phys. Rev. X 4 041009
[14] Senior M L 1979 Prog. Geogr. 3 175–210
[15] Hua C and Porell F 1979 Int. Reg. Sci. Rev. 4 97–126
[16] Kiet H A T, Baek S K, Jeong H and Kim B J 2007 J. Korean Phys. Soc. 51 1812–6
[17] Baek S K, Kiet H A T and Kim B J 2007 Phys. Rev. E 76 046113
[18] Baek S K, Minnhagen P and Kim B J 2011 New J. Phys. 13 043004
[19] Huang K 1987 Statistical Mechanics (New York: Wiley)
[20] Chen Y 2015 Chaos Solitons Fractals 77 174–89
[21] Google Maps https://cloud.google.com/maps-platform/
[22] Korean Statistical Information Service http://kosis.kr/eng
[23] Lee S H, Kim P J, Ahn Y Y and Jeong H 2010 PLoS One 5 e11233
