REVIEW OF ECONOMIC PERSPECTIVES – NÁRODOHOSPODÁŘSKÝ OBZOR
VOL. 15, ISSUE 1, 2015, pp. 3–14, DOI: 10.1515/revecp-2015-0008
Gravity Models of Internal Migration – the Czech
Case Study
Monika Jandová – Tomáš Paleta1
Abstract: In this paper, the models of internal migration flows between regions
(NUTS 3) in the Czech Republic in time series from 1991 to 2012 are tested. The paper
aims to find out how size, distance and economic variables explain migration flows
between Czech regions. Several versions of an extended gravity model were used for
testing, where economic factors which are frequently mentioned in literature on migration
were used as regressors (i.e. average wage, registered unemployment rate, job vacancies
and job vacancies per applicant). Internal migration flow is the dependent variable.
In comparison to the pure gravity model, the extending of models with the economic
variables improves the results of the models only slightly. The results show that the
highest explanatory value of migration is given by models with rates of the variables
tested.
Key words: Gravity model, Internal migration, the Czech Republic, Economic incen-
tives
JEL Classification: J6, J61
Introduction
Migration is an important determinant of economic growth and enhances labour force
utilisation efficiency, both on the international and national level. Results from the national
analysis could contribute to verification of migration theories, to identification of
potential differences between international and national determinants of migration and
can become the groundwork for regional economic policy.
Gravity models are simple but useful tools used in many fields – from geography to
international trade. Besides that, gravity models and their variations are among the
methods most frequently used for determining incentives to move in studies focusing on
migration, both international and internal.2
The work of Lewer and Van den Berg (2007) constitutes the fundamental inspiration for
our gravity model of migration. The authors found out that in conformity with international
trade, migration is driven by attractiveness of individual regions, which is in the
1
Both authors are assistant researchers at the Department of Economics, Faculty of Economics
and Administration, Masaryk University. Contact address: Lipová 41a, 602 00 Brno. E-mail:
jandova@econ.muni.cz and paleta@mail.muni.cz.
2
When we use the term internal migration, it means inter-regional migration.
REVIEW OF ECONOMIC PERSPECTIVES
4
case of migration flows measured by the wage difference. Besides, they assumed that
number of inhabitants was crucial, too, which was justified by two reasons: i) the more
people in a source region, the more people are likely to migrate; and ii) the larger the
population in a target region, the larger the labour market for immigrants. Finally, their
last inspiration with trade is associated with distance (for more details see Lewer and
Van den Berg, 2007).
Many modifications of gravity models can be found in literature. Crozet (2004), for
instance, proved that migration flow between two regions increased first with wage
difference and a probability to find a new job in a target region, second with the size of
a target region, and decreased with the distance between these regions.
In our analysis only migration itself is covered, which means movement of inhabitants
from one region to another that is related to changing a permanent address. Migration
without official changing of a permanent address or commuting is not involved even if
it seems to be a quite important part of flow of inhabitants (especially in case of labour
force).3
We follow our previous work (Jandová and Paleta, 2011; or Paleta and Jandová, 2013)
in our models. Our previous models proved significance of both gravity and economic
determinants of migration but since there was still some space for testing new determinants,
their results were not sufficient enough. Now, the aim is to find out if the extended
gravity model of internal migration with some new variables is capable of explaining
migration flows between Czech regions (NUTS 3) better. Besides, we find out what
kind of variables (which means whether absolute value, difference or ratio) shows the
most statistically important results. The results of our model can answer the question of
what is the most important incentive for a migrant, whether she/he moves on the basis
of absolute values of the variables, their difference, or their ratio. Moreover, results can
be used as a guideline or a methodological tool pro further research in internal migration.
The Starting Point for a Gravity Model
The original version of the Newton's law of gravitation is that there is an attractive force
between two physical objects that is indirectly proportional to the square of the distance
between them, and directly proportional to their mass. In non-physical applications, the
distance can be measured in kilometres, time or price, and the mass is usually replaced
by a number of inhabitants in a region.
In one of our previous papers (see Jandová and Paleta, 2011), we tested solely the gravity
variables based on the Newton’s law of gravity:
log൫‫ܨܯ‬௜௝൯ ൌ ߙ଴ ൅ ߙଵ൫ܱܲܲ௝൯ ൅ ߙଶሺܱܲܲ௜ሻ ൅ ߙଷ൫‫ܶܵܫܦ‬௜௝൯ (1)
in which log൫‫ܨܯ‬௜௝൯ is a logarithm of migration flows from a source to a target region;
ܱܲܲ௝ is a number of inhabitants in a target region; ܱܲܲ௜ is a number of inhabitants in a
3
Its extent could be only estimated, especially on basis of survey questionnaire.
Volume 15, Issue 1, 2015
5
source region; ‫ܶܵܫܦ‬௜௝ is a distance between capitals of regions; ߙ଴ is a constant; and
ߙଵ…ଷ are coefficients.
The results for the period of 1991–2009 were quite satisfactory; both “gravity” variables
proved to be important determinants of internal migration in the Czech Republic. Specifically,
the result was that each kilometre of distance decreased migration flows by
0.635%. The influence of region sizes was proved as well, regardless of the fact if it was
a source or a target region. It was shown that population growth of 100 thousand inhabitants
increased migration flows by 14%.
Our model was inspired by models dealing with international migration. Although some
of the determinants of international migration (e.g. language or culture) are absent in the
decision-making of an “internal migrant”, we suppose that there is no difference in the
fundamental incentives to move. We built our model on the gravity migration model of
Lewer and Van den Berg (2007) mentioned above. Their equation was:
‫݉݉ܫ‬௜௝ ൌ ܽ଴ ൅ ܽଵ൫‫݌݋݌‬௜ ∙ ‫݌݋݌‬௝൯ ൅ ܽଶ൫‫ݕ݈݁ݎ‬௜௝൯ ൅ ܽଷ൫݀݅‫ݐݏ‬௜௝൯ ൅ ܽସ൫‫݇ܿ݋ݐݏ‬௜௝൯
൅ ܽହ‫ܩܰܣܮ‬௜௝ ൅ ܽ଺‫ܱܶܰܥ‬௜௝ ൅ ܽ଻‫ܭܰܫܮ‬௜௝ ൅ ‫ݑ‬௜௝
(2)
in which ‫݉݉ܫ‬௜௝ is a logarithm of immigration to a target region; ‫݌݋݌‬௜ ∙ ‫݌݋݌‬௝ is a multiple
of population sizes; ‫ݕ݈݁ݎ‬௜௝ is a ratio of per capita incomes between a target and a
source country; ݀݅‫ݐݏ‬௜௝ is a distance between regions; ‫݇ܿ݋ݐݏ‬௜௝ is a number of source
region natives already living in a target region; and finally ‫,ܩܰܣܮ‬ ‫ܱܶܰܥ‬ and ‫ܭܰܫܮ‬ are
dummy variables for pairs of regions that share a common language, a contiguous border
and colonial links; ܽ଴ is a constant; and ܽଵ…଻ are coefficients.
In the view of the fact that their model was devised for international migration, we had
to do some important modifications to better fit internal conditions (see Jandová and
Paleta, 2011). First, we removed irrelevant variables such as common language, colonial
relations, contiguous borders and a number of natives from a source region. Moreover,
we replaced Lewer and Van den Berg's equivalent of mass – GDP per capita – with an
average wage because it seemed to be a more important determinant of internal migration
than GDP itself. On the other hand, we were aware of the fact that GDP per capita
and an average wage for a particular region was highly correlated – at least in the Czech
Republic. Our equation in Jandová and Paleta (2011) was formulated as follows:
logሺ‫ܨܯ‬௜௝ሻ ൌ ߚ଴ ൅ ߚଵ൫ܱܲܲ௝൯ ൅ ߚଶሺܱܲܲ௜ሻ ൅ ߚଷ൫ܹ௝/ܹ௜൯ ൅ ߚସ൫‫ܶܵܫܦ‬௜௝൯
൅ ߚହሺܷ௜ሻ ൅ ߚ଺൫ܷ௝൯ ൅ ߚ଻൫‫ܹܲܨ‬௝ െ ‫ܹܲܨ‬௜൯,
(3)
in which logሺ‫ܨܯ‬௜௝ሻ is a logarithm of migration flows from a source to a target region;
ܱܲܲ௝ is a number of inhabitants in a target region; ܱܲܲ௜ is a number of inhabitants in a
source region; ܹ௝/ܹ௜ is a ratio of wages between a target and a source region; ‫ܶܵܫܦ‬௜௝ is
a distance between capitals of regions; ܷ௜ is a rate of unemployment in a source region;
ܷ௝ is a rate of unemployment in a target region; ‫ܹܲܨ‬௝ െ ‫ܹܲܨ‬௜ is a difference between
job vacancies in a target and a source region; ߚ଴ is a constant; and ߚଵ…଻ are coefficients.
This economic model, which was tested on the data set for years 1991−2009, seemed to
be more exact than the gravity one. On the other hand, economic variables did not prove
to be as significant as we had supposed. The growth of the unemployment rate by 1
REVIEW OF ECONOMIC PERSPECTIVES
6
percentage point in a source region increased migration flow from that source region by
2.4%, and contrarily, decrease of unemployment in the target region intensified the
inflow of migrants by 1.9%. In the case of wage ratio between regions, the model
showed that the increase of wage ratio of 1 percentage point increased migration by
27%, however, the reliability level was lower than in the case of previous variables.
Finally, we built our model on Paleta and Jandová (2013), where modifications were as
follows:
logሺ‫ܨܯ‬௜௝ሻ ൌ ߚ଴ ൅ ߚଵ൫ܹ݂݂݀݅௝௜൯ ൅ ߚଶ൫ܷ݂݂ܰ݀݅௝௜൯ ൅ ߚଷ൫‫݂݂ܸ݅݀ܬ‬௝௜൯
൅ ߚସሺ݃‫ݕݐ݅ݒܽݎ‬ሻ.
(4)
In the equation, logሺ‫ܨܯ‬௜௝ሻ is a logarithm of migration flows from a source to a target
region; ܹ݂݂݀݅௝௜ is the average wage in a target region minus an average wage in a
source region; ܷ݂݂ܰ݀݅௝௜ is a registered unemployment rate in a target region minus a
registered unemployment rate in a source region; and ‫݂݂ܸ݅݀ܬ‬௝௜ is a number of job vacancies
in a target region minus a number of job vacancies in a source region; ݃‫ݕݐ݅ݒܽݎ‬
is our combined variable of size and distance; β0 is a constant; and β1-4 are coefficients.
Since distance did not vary within the time and therefore could not be used in the panel
data model directly due to its full collinearity, we constructed a new variable based on
the Newton gravity law. In our case, gravity equals to a population size of a source
region (in thousands) which is multiplied by a population size in a target region and
divided by square of distance (in kilometres) of the regions’ capitals. The coefficient
could be interpreted as an inter-regional “gravity constant”, which must not be mistaken
for “gravity force”. It is impossible in this case of constructed regressor to distinguish
whether the influence is bigger for distance or for size; only joint influence can be tested.
When interpreting the “gravity constant”, it is necessary to bear in mind that the independent
variable is in the log form.
In compliance with our model in Jandová and Paleta (2011), we took into account a
difference in job vacancies between regions. Since we assumed that the difference is
what people motivate to migrate, other variables were modified into a form of differ-
ence.
The results of this model proved that all three economic variables included (i.e. wage
difference, unemployment difference, and difference in job vacancies) were statistically
significant. The question is which other economic determinants of migration are suitable
as a supplement to our model. The problem of long time unemployment suggests that
the probability of finding new job could be more important than we had assumed.
Incentives to Move – Theory and Evidence from Other Models
According to the oldest neoclassical macroeconomic theories migration is caused by
geographic differences in the supply of labour and demand for labour (Masey et al.,
1993).
Volume 15, Issue 1, 2015
7
Generally speaking, migration is determined by push and pulls factors. In accordance
with neoclassical microeconomic theory, Price (1951) stated that people move in order
to improve their standards of living, both economic and social.4
From this point of view,
people who have decided to migrate choose a region associated with the maximum net
benefit from migration (Greenwood, 1974).
Wage difference (see e.g. Massey et al., 1993) is one of the most important pull factors
used in macroeconomic models of migration. This determinant was simultaneously
confirmed by both models we dealt with previously. In the case of the Czech Republic it
is not surprising as e.g. Fidrmuc (2004) and Lux et al. (2006) proved that internal migration
is strongly correlated to average wage. This is why it is unreasonable to reject this
variable from our model.
Push factors are commonly represented by a high rate of unemployment in a source
region and/or a job loss of a migrant. Migrants move from regions with high unemployment
to regions with employment prospects (Fidrmuc, 2004). Moreover, according
to Boeri and Scarpetta (1996), regional unemployment differentials were larger in the
Czech Republic (and other transition countries) than in the EU-12. Both of the determinants
stated above, i.e. a wage level and an unemployment rate, represent possible important
migration incentives but they are not the only ones (see Peker, 2004). Wage
difference as an attractive force can be overweighed by a low probability of finding a
job. Todaro (1968 in World Bank, 2007) explained that a 70 percent wage difference
has a small impact on migration if the probability of finding a job is low (he stated one
to fifty). In the case of the Czech Republic, Srb (1999) found out that migration is correlated,
besides other things, to differences in job opportunities. Paleta and Jandová (2013)
represented the probability of finding a new job by a difference in job vacancies. Unfortunately,
this variable did not reflect number of applicants, which is why modification
of the variable seems desirable.
Distance of regions is another factor determining a willingness to migrate. According to
Price (1949), for instance, we can expect that the number of migrants decreases with
increasing distance, but only in the case of uniform opportunities. On the contrary,
Drobne and Rajar and Lisec (2013) discovered that better transport and working conditions
(i.e. flexible working time or work from home) have a significant positive effect
on our decisions related to commuting instead of migrating. From this point of view,
distance does not seem to play as important role as before but the final impact on migration
is the same.
In total, many authors of empirical studies focused primarily on wage difference and a
rate of unemployment, and a distance as factors determining migration costs (see Borjas,
2010). Besides, we have to admit that there is some evidence confirming the significance
of labour force characteristics (i.e. skills (Dohmen, 2011; or Srb, 1999), risk aversion
(WB, 2007), ownership of houses (Oswald, 1996), age (Vavrejnová, 2004), sex or
4
The neoclassical microeconomic theory of migration presumes that rational individuals decide
to migrate because a cost-benefit calculation (Masey et al., 1993).
REVIEW OF ECONOMIC PERSPECTIVES
8
family status (Dennett and Stillwell, 2008)). These socio-demographic variables were
confirmed as incentives for internal migration in the Czech Republic as well (see
Vajdová, 2002; Fidrmuc and Huber, 2007; or Drbohlav, 1990). A survey on migrants’
incentives described in Vajdová (2002) demonstrated a low significance of economic
factors; it revealed that an effort to find a new job accounted for only about 13–17%.
Unfortunately, socio-demographic data are currently inaccessible from the official resources,
which is why we cannot include them into our model.
Method
As was already mentioned, this paper is a follows-up to our previous work published in
2011 and 2013 (see Jandová and Paleta, 2011; or Paleta and Jandová, 2013). We extend
the data set with new time series (for the year 2012). Unless stated otherwise, Czech
Statistical Office (CZSO) is the source of all the data used. Migration data have been
collected from various issues of statistical yearbooks, regional division CZSO websites,
archives or specific yearbooks focused on migration.5
Unlike the previous papers, we have tested several different models with different variables
this time. Our primary question is whether people take into account values, differences
or rates of involved variables. If one of our models gives substantially better results,
we could make a conclusion about what kind of data people take into consideration.
We have tested all models on panel data for the regions of the Czech Republic. The
length of the time series differs according to availability of the data (see Table 1).
On the basis of theoretical literature and previous research we formulated the following
hypotheses considering the regressors:
1) Wage difference is a factor forcing the migration – the higher the difference (target
minus source), the stronger motivation to migrate. We expect the coefficient in model
results to be positive (both in case of difference and ratio);
2) Unemployment rate difference is a factor preventing migration – the higher the difference
(target minus source), the stronger motivation not to migrate. We expect the
coefficient in the model results to be negative (both in case of difference and ratio);
3) And the same as for wage difference, difference in a number of job vacancies is a
factor forcing the migration. We expect the coefficient in the model results to be posi-
tive.
At the very beginning, we test the “pure” gravity model, looking for the influence of
size (population) and distance only. We use the logarithm of migration flow between the
regions as the dependent variable in all models presented. The logarithm helps us avoid
possible heteroscedasticity problem; in addition to it, it is also better for model interpre-
tation.
The pure gravity model equation is as follows:
5
The collected data set can be provided on request.
Volume 15, Issue 1, 2015
9
logሺ‫ܨܯ‬௜௝ሻ ൌ ߙ଴ ൅ ߙଵሺ݃‫ݕݐ݅ݒܽݎ‬ሻ (5)
Such a starting point allows us to compare the influence of additional variables. In turn
we built up five models with different regressors.
We use the same approach as in Paleta and Jandová (2013) in the first extended model
(6); see above for description of variables. We use inter-regional differences because we
suppose that the difference is what the migrants take into account. The extended gravity
model equation can be formulated as follows:
logሺ‫ܨܯ‬௜௝ሻ ൌ ߚ଴ ൅ ߚଵ൫ܹ݂݂݀݅௝௜൯ ൅ ߚଶ൫ܷ݂݂ܰ݀݅௝௜൯ ൅ ߚଷ൫‫݂݂ܸ݅݀ܬ‬௝௜൯
൅ ߚସሺ݃‫ݕݐ݅ݒܽݎ‬ሻ.
(6)
We used the same variables as in (6) in our next model (7) but in a different form. Now
we did not deal with differences, but rates instead.
The extended gravity model with rates has the following equation:
logሺ‫ܨܯ‬௜௝ሻ ൌ ߯଴ ൅ ߯ଵ൫ܹ‫݁ݐܽݎ‬௝௜൯ ൅ ߯ଶ൫ܷܰ‫݁ݐܽݎ‬௝௜൯ ൅ ߯ଷ൫‫݁ݐܽݎܸܬ‬௝௜൯
൅ ߯ସሺ݃‫ݕݐ݅ݒܽݎ‬ሻ.
(7)
The additional variables are as follows: ܹ‫݁ݐܽݎ‬௝௜ is a rate of wages in a target region and
a source region; ܷܰ‫݁ݐܽݎ‬௝௜ is a registered rate of unemployment in a target region divided
by a registered rate of unemployment in a source region; ‫݁ݐܽݎܸܬ‬௝௜ is a number of
reported job vacancies in a target region divided by a number of reported job vacancies
in a source region; and ݃‫ݕݐ݅ݒܽݎ‬ is the same combined variable of population size and
distance as in the previous model (6).
Our third extended gravity model (8) tested the values of variables for both target and
source region. We involved the ݃‫ݕݐ݅ݒܽݎ‬ variable, too. The equation of extended gravity
model with values is as follows:
logሺ‫ܨܯ‬௜௝ሻ ൌ ߜ଴ ൅ ߜଵሺ݃‫ݕݐ݅ݒܽݎ‬ሻ ൅ ߜଶ൫ܹܽ݃݁௝൯ ൅ ߜଷሺܹܽ݃݁௜ሻ ൅ ߜସ൫‫ܸܬ‬௝൯
൅ ߜହሺܷܰ௜ሻ.
(8)
Quite a lot of variables proved to be statistically insignificant in this model, which was
the reason we omitted them. The following variables were significant: ܹܽ݃݁௝, i.e. wage
in a target region; ܹܽ݃݁௜, i.e. wage in a source region; ‫ܸܬ‬௝, i.e. job vacancies in a target
region; ܷܰ௜, i.e. rate of registered unemployment in a target region; and ݃‫.ݕݐ݅ݒܽݎ‬
We add another variable into the last two models: job vacancies per applicant, first (9)
in a difference, and second (10) in an absolute value form. This variable can be used as
another (and probably better) proxy for probability of finding a new job, than only job
vacancies numbers. Since the time series of the data available includes only 11 years,
we do not incorporate this variable in our previous models (6, 7, 8). Instead of that, we
build two separate models with this variable (9, 10).
logሺ‫ܨܯ‬௜௝ሻ ൌ ߝ଴ ൅ ߝଵሺ݃‫ݕݐ݅ݒܽݎ‬ሻ ൅ ߝଶ൫‫݂݂݅݀ܲܣܸܬ‬௝௜൯. (9)
REVIEW OF ECONOMIC PERSPECTIVES
10
In the model in a form of differences (9), ܹ݂݂݀݅௝௜ and ܷ݂݂ܰ݀݅௝௜ prove to be insignificant.
Only ݃‫ݕݐ݅ݒܽݎ‬ and ‫݂݂݅݀ܲܣܸܬ‬௝௜ are statistically significant, where ‫݂݂݅݀ܲܣܸܬ‬௝௜ is a
number of job vacancies per applicant in a target minus a number of job vacancies per
applicant in a source region.
The equation for the absolute value form is as follows:
logሺ‫ܨܯ‬௜௝ሻ ൌ ߮଴ ൅ ߮ଵሺ݃‫ݕݐ݅ݒܽݎ‬ሻ ൅ ߮ଶሺܷܰ௜ሻ ൅ ߮ଷ൫‫ܲܣܸܬ‬௝൯. (10)
In the last model (10), variables ܹ௝, ܹ௜ and ܷܰ௝ were insignificant, such as was the
number of job vacancies per applicant in a source region (‫ܲܣܸܬ‬௜). The number of job
vacancies per applicant in a target region (‫ܲܣܸܬ‬௝), ܷܰ௜ and again ݃‫ݕݐ݅ݒܽݎ‬ were the only
significant variables.
In order to estimate the results, panel data model with fixed effect was used. To avoid
the problem of autocorrelation and heteroscedasticity we used robust standard errors
(HAC) estimation. We also used log form of dependent variable to deal with heteroscedasticity,
and also because of more convenient interpretation.
The model for the Czech Republic as a whole includes 182 cross-sectional units (migration
flows to and from the same region are omitted), the length of time series varies
between 22 (1991–2012) to 11 (2002–2012) because of (un)availability of any older
data. The summary of the results of our models is displayed in Table 1.
As we can see, the gravity coefficient is positive in all cases, so we can confirm that
gravity matters in the case of internal migration in the Czech Republic. The pure gravity
model, based only on ݃‫,ݕݐ݅ݒܽݎ‬ has quite a high coefficient of determination, and the
coefficient for ݃‫ݕݐ݅ݒܽݎ‬ is (logically) highest in comparison with other tested models.
The high R2
could be explained either by: i) high importance of ݃‫,ݕݐ݅ݒܽݎ‬ or (and) ii) by
stability of migration flow in the period under observation, which gives only little significance
to our variables. The second explanation mentioned is also confirmed by the
results of the other models. Additional variables improve R2
only slightly and the relevance
of the variables tested is quite low. In model (6), for example, the CZK 1000
difference in average wage explains only 0.03 % of migration flow; one percentage
point difference in unemployment rate influences migration only by 1%; and job vacancies
difference of 1000 influences the migration flows only by 0.6%. Such results are
almost the same as in our previous work (see Paleta and Jandová, 2013).
At the first glance, explanatory value of the coefficients in model (7) is much higher.
For example increase of ܹ‫݁ݐܽݎ‬ (i.e. wage rate) by one increases migration by 130%.
But such an increase would require bigger wage differences than those that actually
occur in the Czech Republic (an increase of ܹ‫݁ݐܽݎ‬ from 1 to 2 would require doubling
of wage level in a target region). However, we can say that rate plays a more important
role as the factor of migration than the difference does. The model with values (8) also
has high R2
, and most of its variables are significant. Its explanatory value, nonetheless,
is as low as that of the model (6).
Volume 15, Issue 1, 2015
11
Table 1 Results of the Gravity and Extended Gravity Models for the Czech Republic
Model →
Gravity (5) Diff (6) Rates (7)
Values
(8)
JVAPdiff
(9)
JVAPvalues (10)↓ Independent
variable
Const
5.27
(970)***
5.27
(1669)***
3.92
(15.21)***
5.12
(208.4)***
5.34
(4244)***
5.53
(149.3)***
Wrateji - -
1.3
(5.08)***
Wdiffji(1000) -
0.0003
(2.45) **
- - INS UNrateji
- -
-0.012
(-2.24)**
- - UNdiffji
-
-0.01
(-2.51) **
- - INS JVrateji
- -
0.054
(7)***
- - JVdiffji
-
0,000006
(2.61)***
- - - -
Gravity
0.00029
(10.09)***
0,00019
(6.14) ***
0.00023
(5.4)***
0.00017
(2.7)***
0.00015
(16.65)***
0.000177
(11.28)***
JVAPdiffji - - - -
0.079
(1.95)*
Wt
- - -
0,000038
(4.57)***
- INS
Ws - - -
-0,00002
(-2.37)**
- INS
JVj - - -
0.000015
(12.9)***
- JVi
- - - INS - UNj
- - -
-0.02
(-10.7)***
- INS
UNi - - - INS -
-0.024
(-6.44)***
JVAPj - - - - -
0.17
(4.71)***
JVAPi - - - - - INS
R2 0.94 0.95 0.96 0.96 0.96 0.975
Observation/
time series
4004(22) 3620/20 2912/16 3640/20 1991/11 2002/11
Note: In the case that some variables proved to be insignificant, they were removed from the
model (and market as INS in the table). The sign “-“ (without any value) means that a particular
variable was not used it the particular model. Presented are results of models where insignificant
variable(s) were omitted in the final calculation.
Source: Own calculations (using Gretl software)
The last models (9) and (10) are tested on a shorter time series, so they are not fully
comparable with other models. Nevertheless, their results are interesting. Adding a
number of job vacancies per applicant as a variable makes most of other variables insignificant,
and it also increases the R2
in the case of model (10). This could be interpreted
REVIEW OF ECONOMIC PERSPECTIVES
12
in such a way that probability of finding new job is very important factor for migration,
and is even more important than potentially higher wage.
We tested several gravity models of internal migration in the Czech Republic in our
paper, estimating parameters of our models on the panel data. The aim of the paper was
to find out what economic variables are decisive for inter-regional migration flows in
the Czech Republic and what form (difference, ratio or absolute value) has the most
statistically significant results.
The data set was used with awareness of their limited information capability, when a
part of labour force move is not associated with a change of permanent address, and
furthermore a part of migrants changing a permanent address is not forced by economic
incentives. Up to 22-year-long time series were used (from 1991 to 2012) in order to
find out what form of economic variable migrants take into account. The results of the
model suggest that migration is best explained by ratios of variables. Differences and
values of variables have quite a low explanatory value. It can be interpreted that a migrant
seems to be more interested in a question of “how many times” (represented by a
ratio) than in “how much” (represented by a difference). This finding can be used as a
starting point for an international comparison of internal migration.
Gravity, a combined variable of size and distance, created according to the Newton
gravity laws, explains the migration well, too. Results of the model suggest that economic
variables improve interpretation capability slightly, and a large part of migration
flows seems to be determined by distance of regions, their size or by other than economic
factors that were not covered in this model (due to their inaccessibility).6
Generally, our results are in conformity with theoretical assumptions and with most of
empirical evidence for international migration. But in the case of inter-regional migration
in the Czech Republic, the role of economic variables seems to be weak. The results
confirm conclusions from Fidrmuc (2004) and Fidrmuc and Huber (2007), i.e. that
the economic factors have quite a low impact on internal migration flows in the Czech
Republic. Income (wage) and job vacancies per applicant are the most significant economic
factors of migration. In the case of shorter time series (model 9 and 10) it seems
that the probability of finding a new job is very important factor for migration, and is
even bigger than the potentially higher wage (which is in conformity with Todaro 1968
in World Bank 2007).
The unemployment rate has only small impact on the migration flows. This conclusion
is in contrast with that of Boeri and Scarppeta (1996) who stated that net migration
flows are indeed negatively correlated to unemployment.
Finally, we have to admit that there is an opposite trend, especially in the vicinity of
Prague, where inhabitants change the permanent address and move to the Středočeský
kraj (Central Bohemian Region) without actually changing their jobs. The quite heavy
migration flow from Prague weakens the role of economic factors in the results of our
6
Especially social factors (see e.g. Fidrmuc and Huber, 2007).
Volume 15, Issue 1, 2015
13
models. We can expect a stronger real impact of economic variables than expressed in
the results of our model.
References
BOERI, T., SCARPETTA, S. (1996). Regional Mismatch and the Transition to a Market
Economy. Labour Economics 3, pp. 233-254.
BORJAS, G. J. (2010). Labor Economics. Boston: McGraw Hill.
CROZET, M. (2004). Do migrants follow market potentials? An estimation of a new
economic geography model. Journal of Economic Geography. 4, pp. 439–458.
Database of the Czech Statistical Office. Available from Internet: www.czso.cz.
DENNETT, A., STILLWELL, J. (2008). Internal migration in Great Britain – a district
level analysis using 2001 Census data. WP 01/08, University of Leads.
DOHMEN, T. (2000). Housing, mobility and unemployment. Discussion Paper No. 210.
DRBOHLAV, D. (1990). Migrační motivace, regionální a sídelní preference
obyvatelstva. Sociologický časopis. 26(5), pp. 358–374.
DROBNE, R., RAJAR, T., LISEC, A. (2013). Dynamics of Migration and Commuting
to the Urban Centres of Slovenia, 2000-2011. Geodetski vestnik. 57/2, pp. 333–353.
FIDRMUC, J. (2004). Migration and regional adjustment to asymmetric shocks in
transition economies. Journal of Comparative Economics 32, pp. 230–247.
FIDRMUC, J., HUBER, P. (2007). The willingness to migrate in the CEECs. Evidence
from the Czech Republic. WIFO Working Papers 286.GREENWOOD, M. J. (1974).
Research on Internal Migration in the United States: A Survey. Journal of Economic
Literature. 13(2), pp. 397–433.
JANDOVÁ, M., PALETA, T. (2011). Vnitřní migrace, In: Slaný, A. et al. (eds.),
Konkurenceschopnost, růstová výkonnost a stabilita české ekonomiky (pp. 41–56). Brno:
Masarykova univerzita.
LEWER, J. J., BERG, H. van den (2007). A Gravity Model of Immigration. Economic
Letters, 99, pp. 164–167.
LUX, M. et al. (2006). Analýza opatření bytové politiky směřujících k podpoře flexibility
práce v ČR. 1. díl, teoretický úvod. Praha: Sociologický ústav AV ČR.
MASSEY, D. S. et al. (1993). Theories of International Migration: A Review and Appraisal.
Population and Development Review. 19(3), pp. 431–466.
OSWALD, A. (1996). A Conjecture on the Explanation for High Unemployment in the
Industrialized Nations: Part I. Warwick Economic Research Paper No. 475.
PALETA, T., JANDOVÁ, M. (2013). Incentives to move: a model of inter-regional
migration in the Czech Republic, In Machová, Z., Uhrová, N. Economic Policy in the
European Union Member Countries (pp. 270–282). Ostrava: VŠB - Technical university
Ostrava, 2013.
REVIEW OF ECONOMIC PERSPECTIVES
14
PEKER, K. (2004). The causes and results of internal migration from rural areas: case
of Eastern Anatolia. Agricultural Economics. 50(10), pp. 471–476.
PRICE, D. O. (1949). Distance and Direction as Vectors of Internal Migration. Social
Forces. 1, pp. 48–53.
PRICE, D. O. (1951). Some Socio-Economic Factors in Internal Migration. Social
Forces. 4, pp. 409–415.
SRB, V. (1999). Vnitřní stěhování 1991–1997 podle důvodů migrace. Urbanismus a
územní rozvoj. 3, pp. 8–14.
VAJDOVÁ, Z. (2002, February 6). Kdo, kam a proč se v minulosti stěhoval? Postoje k
bydlení v ČR 2001. Retrieved September 11, 2010, from
http://seb.soc.cas.cz/postoje2001/prezentace/seb_vajdova_drahy.pdf
VAVREJNOVÁ, M. (2004). Mobilita pracovní síly před a po vstupu ČR do EU. Acta
Oeconomica Pragensia. 12(3), pp. 195–218.
World Bank. (2007, January 8). Overview of Migration Trends in Europe and Central
Asia, 1990–2004. Retrieved August 29, 2010, from
http://siteresources.worldbank.org/INTECA/Resources/257896-
1167856389505/Migration_Chapter3.pdf.