Please limit the report to max. 3-5 pages, including tables and figures and use the following structure as much as possible.
Name and last name
Tim Vetter
Project title
Influence factors on the immigration decision on the municipal level: A multilevel model approach
Abstract (max 300-500 words)
The aim of the project is to identify variables that explain the volume of immigrant inflows into municipalities. This shall be achieved through multilevel linear regression analysis. On the first level of the model, the impact of individual and household variables on the individual’s migration decision are analysed. On the second level of the model, variables containing aggregated data on the municipal level are used to account for the influence of municipal macro structures on the individual’s migration decision. Variables on the second level of the model thereby constitute an initial moving probablilty that differs from municipality to municipality.
At CED, I decided to analise influence factors on the determinants of immigration of unemployed persons into 42 Catalan municipalities. Catalunya is an interesting case as, following the financial crisis (2009), unemployment was very compared to other European countries like Germany. At the same time, unemployment rates varied significantly between municipalities (see Table 1). If municipal unemployment rates have an impact on the individual descision to migrate into a municipality, the inclusion of the municipal unemployment rate into a regression model is more promising for understanding varying immigration numbers into municipalities than a model that neglects these structural factors, e.g by relying solely on individual characterisitcs. Data relevant for the first level of the model (individual characteristics of unemployed immigrants/stayers) as well as data for the second level of the model (socio-economic characteristics of the Catalan municipalities) can be extracted and, in the case of aggregated variables, calculated from IECM / IPUMS database. Below, IECM / IPUMS data as well as some very preliminary test calculations are presented.
Introduction and motivation of visit
On previous research (Vetter 2015) I analysed the impact of several macro-level variables on the variation of immigrant inflows into Bavarian municipalities. The analysis would have benefitted from an analysis that uses micro-level data, i.e., by taking characteristics of individual immigrants into account. Gaining access to the relevant micro-data allows for such richer analysis.
Scientific objectives of visit
Not much empirical work explains migration patterns at the sub-national level, e.g. Buch et al. (2013), Deas/Hincks (2014), and Sandell (2008). Also, migration is a “multifactorial and multidimensional phenomenon“(Piché 2013: 157). It can be assumed that the macro structure (a low unemployment rate of a municipality) has an impact on the individual’s migration decision However, there are also many characteristics influencing the migration decision on the individual or household level that should be taken into consideration, such as the age of the individual or family ties. Hierarchial (or multilevel) modelling considers micro and macro variables systematically and could help to gain new insights on the determinants of immigration into municipalities (Chi/Voss 2005; Matthews/Parker 2013).
Reasons for choosing research infrastructure and datasets/surveys/...
There are several reasons why I chose to conduct my work at CED with IECM/IPUMS data. Firstly, CED offers access to data of various countries and years, containing many variables. This data is needed to conduct my work that relies heavily on microdata. Here, I had the possibility to experiment with different variables, sample sizes, country selections. Furthermore, researchers at CED have an expertise in migration and demography and are experienced in manipulating and working with microdata. All this provided an excellent starting point. for my research activities.
Activities during your visit (research, training, events, ...)
Besides working on my project, I also had the opportunity to visit a lecture given by Mariona Lozano Riera on domestic division of labour. Furthermore, I could visit research seminars (e.g, Presentation given by Carles Millàs i Castellví on immigration into the Catalan municipality Olesa de Montserrat from a long-term point of view 1581-1930).
Method and set-up of research
For the multivariate regression approach, a logit regression model is run for the Level 1 data. The dependent variable is the migration decision of the individual (did the person move to the municipality within the last 12 months – Yes/No).
At Level 2, the intercept and each coefficient are used as dependent variables, which can be explained by independent variables at that level. The intercept and the coefficients can vary significantly across the different municipalities. The unique characteristics of each municipality provide and determine a unique “initial” move probability for the Level 1 model. For an introduction to hierarchial modelling with R, see Bates (2010) or Bates et al. (2015).
The dependent variable is taken from the IECM and IPUMS census microdata database (did the person immigrate into a Catalan municipality within the last 12 months). The sample (Spain 2011) contains 604,627 observations collected from respondents whose residence was in one of the 42 Catalan municipalities in 2011. I don’t differ between international immigration (immigration from abroad into the Catalan municipality) or internal migration (immigration from within Spain into the Catalan municipality), as the focus of the analysis is on the migration into the municipality. This covers internal migration as well as international migration into the municipality.
For my test calculations, I decided to exclude persons employed and inactive persons from the sample. I decided to focus on influence factors on the immigration decision of unemployed persons aged 20-64 to exclude family migration to a certain extent. 66,341 respondents aged 20-64 were unemployed, of whom 1,740 persons (2.6% on all unemployed persons) had immigrated into a Catalan municipality within the last 12 months.
On the first (individual) level of the model, the following variables were extracted from IECM / IPUMS database: age, nationality, and marriage status (married/not married) with or without children. These variables are included in the regression model as individual cost-benefit calculations influence the migration decision. Other variables I also worked with were variables concerning skills levels, sex, or data on the Spanish province level. For a summary of factors determining the immigration decision, see e.g. Massey et al. (1993) or Bodvarsson/Van den Berg (2013).
For unemployed persons, I assume that one influence on immigration into a municpality on the second (municipal) level of the model is the municipal unemployment rate. This is because high unemployment rates discourage immigrants. Also, the share of persons living in their own houses/flats as a share on the municipal population is added to include a housing market variable. It can be assumed that unemployed immigrants are less likely to buy own property and therefore look for rental housing. If the share of persons living in their own houses is comparatively low, a comparatively high share of housing space is available to rent. Municipalities with a comparatively low share of persons living in their own houses on the total population might therefore be more attractive for unemployed immigrants.
All indicators are taken or calculated from IECM and IPUMS census microdata database. I restricted my analysis to integrated variables contained in the IECM and IPUMS census microdata database. Therefore, it should be possible to conduct the same calculations for non-Spanish (or non-Catalan) municipalities or samples from other years at a later time.
Project achievements during visit (and possible difficulties encountered)
Archievements:
Generally, besides gaining experience in working with micro data and hieracial modelling, it was very fruitful to exchange views with researchers and Ph.D. students at CED. I gained insight into Spanish and Catalan migration patterns, which in part, did not match my prior expectations. These experiences will help to set up a more sophisticated research set-up.
It is often difficult to obtain data on the sub-national level. Especially on the municipal level, working with micro data proved to be very useful, as the data can be adaptated freely.
Difficulties:
In the German case, the latest data contained in the IECM/ IPUMS database is from 1987. It was therefore not possible to conduct my research on the geographical level of NUTS3 for Germany (which corresponds to the German municipal level). It would have been interesting to see if the findings of my previous research would have been confirmed when working with microdata. However, it was sufficient for the purpose of my visit to use Spanish census data from 2011, as the focus of my work was manly to to gain experience in working with micro data and to work with the mixed model regression methodology.
One main advantage of analysis of immigration of unemployed persons is that all mentioned variables (which are however not exhaustive for analysing these migration streams) can be taken or aggregated from IECM and IPUMS datasets. Analysis of other groups than unemployed would demand a more diverse dataset including more variables extracted from different sources. For instance, analysis of immigration decisions of employed persons should also include variables like municipal wage rates that are not contained in the Spanish sample that was extracted from the IECM/IPUMS database. Also, more detailed housing market variables would be useful to analise initial migration probability into a municipality (e.g. level of rental obligations on the municipal level). It is therefore necessary to include variables that need to be extracted from other sources. It is however difficult to combinine data from different sources due to varying geographical classifications on the sub-national level, depending on the source of the data. E.g., in contrast to Germany, NUTS 3 classification does not correspond to the municipal level in the Spanish case. It was therefore not resonable to include data on the NUTS3-level that could have been extracted online from Eurostat Database.
Preliminary project results and conclusions
Municipal unemployment rates ranged from 16.95% (in San Cugat del Vallés) to 34.85% (in Blanes) in 2011 (see Table 1). The share of persons living in their own homes ranged from 12.99% (in Castelldefels) to 40.19% (in Mataró). The Gini index indicates that the assumption of varying initial migration probabilities is valid, as it is not zero and does therefore not indicate equal distribution. All Catalan municipalities were assigned to one of four unemployment groups (from low unemployment to high unemployment) as well as to one of four groups indiciating the share of persons living in their own homes.
Table 1 Overview of random effects / level 2 variables
Variable (level 2, 42 observations) / Min / Median / Mean / Max / Gini indexMunicipal unemployment rate (% on labour force, age 15-64) / 16.95 / 25.12 / 25.19 / 34.85 / 0.08
share of persons living in their own homes (% on total population) / 12.99 / 19.83 / 20.17 / 40.19 / 0.16
Source: IECM/IPUMS, own calculations
Level 1 variables are shown in Figure 1. The figure shows that immigration is more likely if an unemployed person is young, of foreign citizenship, and not married as well als childless.
Figure 1 Immigration of unemployed persons into Catalan municipalities by age, nationality and marriage status
Source: IECM/IPUMS, own calculations
Preliminary hierarchial regression results indicate that immigration into a Catalan municipality is less likely for older unemployed than for younger unemployed persons. Unemployed who are not married and have no children are more likely to immigrate into a Catalan municipality. Unemployed with Spanish citizenship are also less likely to decide to immigrate into a Catalan municipality. In model 2, model 1 is supplemented by the municipal unemployment rate as a second level 2 variable (or second random effect). Surprisingly, model 1 shows a better fit than model 2 (lower AIC and BIC values – see Table 2).
Table 2 Comparison of models
Df / AIC / BIC / logLik / deviance / Chisq / Chi Df / Pr(>Chisq)model 1 / 5 / 14826.81 / 14872.32 / -7.408.406 / 14816.81
model 2 / 6 / 14828.76 / 14883.37 / -7.408.379 / 14816.76 / 0.05217589 / 1 / 0.8193194
Source: IECM/IPUMS, own calculations
There is still much methodological refinement to do (e.g. allowing for random slopes within municipalities, definition of groups). Furthermore, more variables need to be included (e.g. variables on education). Model 1 contains the above mentioned level 1 variables (fixed effects) age, nationality, marriage status, and the share of persons living in their own homes as a level 2 variable (random effect).
Outcomes and future studies
My visit at CED was very fruitful. The whole CED staff was very forthcoming and provided support for all administrative and scientific questions. Initially, I planned to publish a working paper or a scientific article. After my visit at CED, I also consider to focus on the topic more extensively within the scope of a PHD thesis.
References
Bates, Douglas M. (2010): lme4: Mixed-effects modelling with R. Springer, Berlin. Online: http://lme4.r-forge.r-project.org/lMMwR/lrgprt.pdf.
Bates, Douglas; Maechler, Martin; Bolker, Ben; Walker, Steve (2015). Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software, 67(1), 1-48.
Bodvarsson, Örn B.; Van den Berg, Hendrik (2013). Springer, New York.
Buch, Tanja; Hamann, Silke; Niebuhr, Annekatrin; Rossen, Anja (2013): What Makes Cities Attractive? The Determinants of Urban Labour Migration in Germany. Urban Studies 1-19, 2013.
Chi, Guangqing; Voss, Paul (2005): Migration Decistion-making: A Hierarchial Regression Approach. The Journal of Regional Analysis & Policy 35 (2), p. 11-22.
Deas, Iain; Hincks, Stephen (2014): Migration, Mobility and the Role of European Cities and Regions in Redistributing Population. In: European Planning Studies 22 (12), S. 2561-2583.
Massey, Douglas S.; Arango, Joaquin; Graeme, Hugo; Kouaouci, Ali; Pellegrino, Adela; Taylor, J. Edward (1993): Theories of International Migration: A Review and Appraisal. Population and Development Review 19 (3), pp. 431-446.
Matthews, Stephen A.; Parker, Daniel M. (2013): Progress in Spatial Demography. In: Demographic Research 28 (10), p. 271-312.
Piché, Victor (2013): Contemporary Migration Theories as Reflected in their Founding Texts. In: Population 68 (1), S. 141-164.
Sandell, Rickard (2008): Immigration and cumulative causation: Explaining the ethnic and spatial diffusion of Spain's immigrant population 1997-2007. Madrid Insitute for Advanced Studies Social Science Department, working paper series 2008/12.