Network models for analyzing international migration

The level of education in origin country is presented in interaction with diaspora. Intuition behind this idea is the following. The effect of diaspora on bilateral migrant stock is the sum of its direct and interaction effects. The interpretation of coefficient before the migrant stock variable is changing in this way, i.e. represents now the effect of diaspora on migrant stock when the interaction variable () equals to 0. The overall influence of diaspora will be the sum of its direct effect and the effect of interaction with education variable (). It is expected that diaspora should play less significant role for migrants from more educated countries. Thus, the coefficient before this interaction variable is expected to be negative.

Another interaction term is magnitude of conflict in origin country and intercountry distance. The intuition behind this hypothesis is that conflict in origin should increase the role of distance for migrants. As conflicts induce migrants to leave country and move to the most proximate one with stable situation. Also, the costs related to journey to new country are higher for migrants living in conflict area compared to those originating from the country without conflicts.

According to [4] common language in origin and destination country of migrants should increase bilateral migration flows, as it favors faster integration of migrants into the host society. The dummy variable for common official language from CEPII data set () is included. It equals 1, if countries have the same official language.

Income inequality is the rising global issue during the last decade. The mechanism of interaction of income inequality and international migration is not straightforward. On the one hand, high-skilled labor force aimed to maximize the return on their human capital is aimed to move to those countries with huge concentration of opportunities, high technological development. These migrants are motivated to increase their income and become a part of a small percent of population with the highest income in a new country. On the other hand, migrants that aim to benefit just from moving to country with higher income do not necessary need high income inequality in a new country, as it is related for them with the risk to join the share of people with lower wages. Thus, they will prefer to choose a country with a smaller level of income inequality. In our model, we will test the effect of income by Gini coefficient. Gini coefficient measures the deviation of real income distribution from the perfectly equal income distribution in the economy of the specific country. The destination country Gini index is included to our model.

The model will explore the influence of country pair effects, i.e. diaspora, distance and common language, economic factors of origin and destination (GDP and income inequality), social and political characteristics (conflicts and education level). Variables and descriptive statistics are listed in Table 3.1.

Table 3.1. Descriptive statistics

The next step is the estimation of panel data regression. Our data have a panel structure, where each observation represents the data on country pair at each period. Following [23] there are three main approaches of estimation the panel data regression: pooled regression, fixed effects (FE) and random effects (RE) models.

Pooled model assumes that the difference between each country observation is constant for all origin and destination countries. In this way pooled model ignores country effects that cannot be explained by observed variables. Including the fixed effects for origin and destination countries can bridge this gap and account for unobservable country characteristics. In model (3.1) and denote the origin and destination country effects respectively that are constant throughout the period. For example, our model does not include variables for migration policy, openness to foreigners in destination countries, climate and cultural characteristics of origin and destination, which can be captured by origin and destination effects.

Random effect model (RE) assumes that individual differences by countries are random and do not change over time [3], i.e. and in this case represent the random error for each origin and destination country that are constant for all . It is interesting to explore, how the different panel data approaches will estimate the influence of economic and political factors of origin and destination countries on bilateral migrant stock.

3.3 Results of estimation of panel data models of international migration

Models with different variable combinations have been estimated for 214 countries for the period from 1990 to 2015 (by 5 year intervals). Pooled, fixed and random effect models have been estimated on 20878 observations of dependent variable.

First, the pooled model has been estimated. Model appears to be significant according to F-statistics on 1 % level. However, the coefficient before dummy for common language is insignificant. In this way, other coefficients of model cannot be interpreted. The same problem is with the random effect model: coefficient before Ln GDP per capita at origin has been insignificant.

The model with all coefficients significant at the level 1 % is fixed effects model (FE). The following equation is obtained. Standard errors are placed in parenthesis.

(3.2)

Certain country characteristics introduced by the model (3.2) can be highly correlated. For example, education index and GDP of origin country. High correlation among the explanatory variables (multicollinearity) can lead to biased and inconsistent estimates of the regression coefficients [1]. To address this issue pairwise correlation matrix is obtained for explanatory variables (Appendix 2). The maximal value of correlation is 0,6191 (among interaction variable of diaspora and education and GDP of origin country). Thus, we can continue the analysis of our model. The next issue that is needed to be addressed is the error term specification. In model (3.2) White heteroscedasticity consistent standard errors are included to make OLS estimations consistent.

When the basic properties of regression estimates are considered, it is possible to interpret the obtained coefficients in the model (Table 3.2).

Table 3.2 Estimates of panel data model of international migration

Standard errors in parentheses

^* p < 0.05, ^** p < 0.01, ^*** p < 0.001

According to FE model estimates (column 2) the highest influence on bilateral migrant stock is caused by diaspora at destination country. The coefficient before diaspora is 0,869, which means that migrant stock from certain country in the country with large diaspora should be 86,9 % higher compared to the one with the smallest diaspora, ceteris paribus. Interestingly, the level of education in origin country reduces the pull effect of diaspora in destination by 6 %. This finding indicates that migrants from countries with lower education level are more dependent on the existing network from their origin. However, even for countries with the highest level of education diaspora plays a very important role and accounts for 80,9 % changes in bilateral migrant stock.

Another pull factor of origin country is GDP per capita. The increase in per capita GDP at origin by 1 % will lead to 3,96 % increase in bilateral migrant stock. However, the effect of GDP per capita turned out to be modest. The effect of GDP per capita at origin should be two-sided according to previous studies [4,17]. It should be positive until it reaches a certain level, as the higher is GDP per capita at origin, more potential migrants will be able to finance their journey. Then it becomes negative, as difference in countries' welfare appears to be smaller and the less potential migrants are motivated to leave their home country. Our model has captured only the negative effect of GDP per capita at origin meaning that higher GDP per capita at origin country reduces the bilateral migrant stock by 15,9 %.

The effect of distance in our model is explored separately and in interaction with intensity of conflict at origin country. More distant countries are characterized with 11,7 % lower migrant stock compared to those most proximate. The influence is quite modest. Possible explanations of this result are the following. First, international transportation system became more developed and differences between long distant journey and short one became less important for international migrants. Second, approach with measurement intercountry distances as distances between the country capitals may have certain shortcomings and lead to biased estimates of distance effect on international migration.

Interesting fact is obtained by including the magnitude of conflict variable in interaction with distance. Negative coefficient before this interaction variable indicates that higher magnitude of conflicts at origin country in previous period increases the effect of distance on migrant stock.

Countries sharing common official language are expected to have 9 % higher migrant stock compared to those having different official languages. However, it is important to mention that common official language is only a part of overall influence of language on international migration. Also, having languages from the same group can favor the increase in migrant stock [4], as language is the indicator of cultural proximity.

The indicator of income inequality at destination () appeared to be insignificant in all model specifications. This result is possibly explained by the fact that among countries with high Gini coefficient there is a large share of Latin American countries (Colombia, Brazil). On the contrary, European countries have lower values of GINI index. At the same time Europe attracts large amount of migrants. In this way, difference in Gini index does not explain the difference in migrant population.

Conclusion

To sum up, the following results have been obtained in Chapter 3. The data set on the variety of economic and political factors of origin and destination countries has been considered for the analysis of determinants of international migrant stock. The panel data models were studied to find the most important factors influencing international migration. The following findings have been obtained. First, it is shown that fixed effect model better describes the process of international migration. Second, along with the influence of traditional factors (GDP per capita, intercountry distance, diaspora) the role of education level and magnitude of conflicts at origin is examined. It is shown that higher education in sending country reduces the effect of diaspora and conflicts at origin increase the role of distance on international migrant stock.

Conclusion

International migration can be modeled in various ways. Extensive amount of works study the international migration flows on country level and analyze the causes of their emerging. Gravity models can help to benefit from longitudal panel data and observe the determinants of bilateral migrant flows. Next, network analysis allows to represent all countries as a system and consider the migration flows between any two countries as an imprescriptible part of the international migration flows in the whole network.

The network analysis has been applied to annual international migration flows. The influence of countries in the network has been evaluated by classic and new centrality indices. Classic centrality indices form the basic understanding of the migration networks and find countries with considerable inflows and outflows of migrants. SRIC and LRIC indices find new influential countries in the international migration network, precisely, those with outflows considerable for the receiving country population. This result is accomplished by accounting for indirect influence and group influence in the network and host country population.

Econometric model has been analyzed on bilateral international migrant stock data. Different model specifications have been considered. The analysis confirms the importance of traditional determinants of international migration (diaspora, income and distance). However, new interesting findings are obtained as well. The influence of distance turns out to be larger for those origin countries with higher magnitude of conflicts. The role of diaspora appears to be lower for more educated origin countries.

Combining the network analysis with econometric modeling gives us the comprehensive view on the international migration process. The countries with different type of influence in the network of migration flows have been detected and certain motives influencing the decisions of migrants are found.

References

1Айвазян С. А., Мхитарян В. С. Прикладная статистика и основы эконометрики: Учебник для вузов. - 1998.

2Вакуленко Е. С., Мкртчян Н. В., Фурманов К. К. Моделирование регистрируемых миграционных потоков между регионами Российской Федерации //Прикладная эконометрика. - 2011. - №. 1 (21).

3Магнус Я. Р., Катышев П. К., Пересецкий А. А. Эконометрика. - 2004.

4Adsera A., Pytlikova M. The role of language in shaping international migration //The Economic Journal. - 2015. - Т. 125. - №. 586. - С. F49-F81.

5Aleskerov F.T. Power indices taking into account agents' preferences // Mathematics and Democracy, Berlin: Springer, pp. 1-18, 2006

6Aleskerov, F., Andrievskaya, I., & Permjakova, E. Key Borrowers Detected by the Intensities of Their Short-Range Interactions, Higher School of Economics Research Paper, 2014

7Aleskerov F., Meshcheryakova N., Shvydun S. Centrality measures in networks based on nodes attributes, long-range interactions and group influence //arXiv preprint arXiv:1610.05892. - 2016.

8Aleskerov F., Meshcheryakova N., Rezyapova A., Shvydun S. Network analysis of international migration. - 2016.

9Bavelas A. Communication patterns in taskЃ]oriented groups //The Journal of the Acoustical Society of America. - 1950. - Т. 22. - №. 6. - С. 725-730.

10Beine M., Bertoli S., FernбndezЃ]Huertas Moraga J. A practitioners' guide to gravity models of international migration //The World Economy. - 2015.

11Bodvarsson, Ц. B., Simpson, N. B., & Sparber, C., 2015. Migration Theory // Handbook of economics of international migration. Volume 1A The Immigrants (1st ed., Vol. 1). Elsevier Inc.

12Bonacich P. Technique for analyzing overlapping memberships //Sociological methodology. - 1972. - Т. 4. - С. 176-185.

13Brin S., Page L. The anatomy of a large-scale hypertextual Web search engine //Computer networks and ISDN systems. - 1998. - Т. 30. - №. 1-7. - С. 107-117.

14Brown, Richard P.C., Jimenez-Soto, E., 2015. Migration and Remittances // Handbook of the Economics of International Migration (1st ed., Vol. 1B). Elsevier B.V.

15Burger M., Van Oort F., Linders G. J. On the specification of the gravity model of trade: zeros, excess zeros and zero-inflated estimation //Spatial Economic Analysis. - 2009. - Т. 4. - №. 2. - С. 167-190.

16Collier P. Exodus: How migration is changing our world. - Oxford University Press, 2013.

17Collier P., Hoeffler A. Migration, Diasporas and Culture: an Empirical Investigation //University of Oxford, Department of Economics Economics Series Working Papers. - 2014. - №. WPS/2014-27.

18Davis K., D'Odorico P, Laio F, Ridolfi L., 2013. Global Spatio-Temporal Patterns in Human Migration: A Complex Network Perspective // PLoS NE8(1): e53723.

19Fagiolo G., Mastrorillo M. International migration network: Topology and modeling //Physical Review E. - 2013. - Т. 88. - №. 1. - С. 012812.

20Freeman C., Centrality in social networks: Conceptual clarification //Social networks. - 1979. - Т. 1. - №. 3. - С. 215-239.

21Gonзalves B. et al. Immigrant community integration in world cities //arXiv preprint arXiv:1611.01056. - 2016.

22Goodman L.A., Kruskal W.H. Measures of Association for Cross Classifications // Journal of the American Statistical Association Vol. 49. No268. P.732-764, 1954

23Greene W. H. Econometric analysis. - Pearson Education India, 2003.

24Izquierdo M., Jimeno J. F., Lacuesta A. Spain: from immigration to emigration?. - 2015.

25Lee, E. A Theory of Migration. // Demography, 3(1), 47-57, 1966

26Mayer T., Zignago S. Notes on CEPII's distances measures: The GeoDist database. - 2011.

27Mincer, J. Family migration decisions. // J. Polit. Econ. 86 (5), 749-773, 1978.

28Polachek, S. W., & Horvath, F. W. A Life Cycle Approach to Migration: Analysis of the Perspicacious Peregrinator. // Research in Labor Economics 35th Anniversary Retrospective (Vol. 35, pp. 349-395), 2012

29Ravenstein, E. G., The Laws of Migration // Journal of the Royal Statistical Society, 52(2), 241-305, 1889.

30Sjaastad, L. The costs and returns of human migration. // J. Polit. Econ. 70, 80-93, 1962.

31Smith A. An inquiry into the nature and causes of the wealth of nations (ed. RH Campbell, AS Skinner, and WB Todd), 1976.

32Tinbergen J. Shaping the world economy: suggestions for an international economic policy. // №. HD82 T54., (The Twentieth Century Fund, New York), 1962

Размещено на Allbest.ru