A centre-oriented survey among EU citizens in Germany: the experience with Bulgarians in Hamburg. Vesela Kovacheva

Transcription

1 A centre-oriented survey among EU citizens in Germany: the experience with Bulgarians in Hamburg Vesela Kovacheva Paper prepared for the ESRA conference July 2013, Ljubljana Table of content 1 Introduction: surveying persons with migration background Survey design: the choice of location sampling Data collection: pretest and field work in Hamburg Sample description: weighting, quality and estimates Concluding remarks: chances and challenges of the centre-oriented survey References Introduction: surveying persons with migration background The distinction between EU and non-eu citizens has gained in importance particularly driven by migration policies at the European level, targeting separately EU and non-eu citizens. Currently, about 500 million persons possess the citizenship of an EU member state and are thus citizens of the European Union too. The category of EU citizens is, however, quite diverse, consisting of persons who obtained EU citizenship in different ways at birth, through acquisition of the citizenship of an EU member state or as a result of political developments like EU enlargements. In the individual member states the distinction based on EU citizenship appears to be too broad for analysis and the member states have preferably employed criteria such as national citizenship and/or country of birth. In Germany, with the introduction of the new statistical category persons with migration background in 2005, the understanding of migrant population has radically changed. In contrast to the classical understanding of the migrant population as encompassing persons with foreign citizenship, the concept of migration background defines a much broader category, taking into account citizenship, country of birth and origin of parents. Persons with migration background are foreign citizens, naturalized persons and German citizens with at least one parent born abroad. Therefore, results from surveys among foreign nationals are not representative for the migrant population in Germany (Galonska et al. 2004) and surveys increasingly aim at capturing the population with migration background. Smaller migrant groups, which do not belong to the main groups with migration background like those with Turkish, Russian, Polish, Italian and Spanish background, can be considered hard to reach populations. Hard to reach means that the population of interest has relatively low numbers, the members of the population are hard to identify, there is no sampling frame or only a very incomplete one, the persons concerned to not wish to disclose that they are members of this population of interest, the behaviour of the population of interest is not known (Marpsat and Razafindrasima 2010: 4).

2 2 Surveying persons with migration background in a representative way often appears to be a difficult task due to a lack of data sources for building a proper sampling frame. Survey samples can be drawn on the basis of administrative registers, telephone books (onomastic sampling), censuses and large-scale surveys. Main sources for sampling drawing in Germany are the local population registers, the Central Register of and the Microcensus in combination with the Register of Foreign Nationals. Some surveys have employed onomastic sampling by screening telephone books for culture-oriented names such as Turkish and Russian names. However, these sources have certain peculiarities that make them inappropriate to survey individual rather small EU groups with migration background. The local population registers captures citizenship but not country of birth, so that naturalized persons cannot be identified and included in the sample. They do not include unregistered persons and particularly in the case of intra-european free movement of persons, there is evidence that some mobile EU people move without registering in the destination country, which may lead to undercounting. Likewise, the Central Register of Foreigners covers only foreign nationals. Moreover, in the case of EU citizens the registration in the Central Register of Foreigners has been challenged by an Austrian citizen who felt discriminated on the grounds of nationality due to the storage and processing of his personal data in the Register. Yet, it is not clear how Germany is going to comply with the decision of the European Court of Justice. It may be expected that the storage of data on EU citizens would be restricted, thus losing its relevance for surveying EU citizens. No census has been carried out in Germany since 1987 but the annual Microcensus survey providing information about 1 per cent of the German households is a main source for extrapolations to the total population. The Microcensus survey captures the category migration background since 2005 and thus constitutes a solid base for analysis of the main countries of origin (Turkish, Polish and Russian) and broader categories such as EU-8 member states. However, it fails at including enough cases of smaller countries of origin, thus making sample drawing, analysis and estimates based on Microcensus for them hardly possible. Universal broad population surveys such as the Socio-Economic Panel (SOEP) do not qualify for including small groups in a representative way (Frick 2004, p. 8). Against this background, the need for a survey designed for an individual group of interest with certain migration background becomes apparent. Migration from the EU member state Bulgaria has substantially increased since 2007 and the country became a main sending country for Germany, so that migration has drawn much political, public and media attention. In spite of the increasing inflows and stock of migrants, the group is still small. Applying the abovementioned criteria to the population with Bulgarian migration background, the lack of a list of the units of the population (sampling frame) and the small number in the population makes an argument for considering it hard to reach. The Bulgarian migrant group encompasses not only Bulgarian citizens but also naturalized persons who possessed Bulgarian citizenship in the past and German citizens with at least one parent originating from Bulgaria. All of them have in common the EU citizenship status gained as a result of individual naturalization before 2007 or the EU accession of Bulgaria in In the doctoral thesis the meaning of EU citizenship by the most recent EU citizens is explored. Reaching them in the framework of a survey based on the population register or Central Register of Foreigners is hardly possible and would lead to underrepresentation of subgroups like naturalized and unregistered persons. The paper seeks to address this challenge and presents an innovative sampling frame a centre-based survey and its implementation for surveying persons of Bulgarian origin in

3 3 Hamburg. 1 The survey consists in interviewing persons at selected migrant-oriented meeting points such as commercial, religious and educational centres. Main assumption is that each person visits at least one of the selected centres, so that a sample is drawn at random at them (random sampling). The bias of the sample is addressed ex post by weighting, which is based on a calculation of the likelihood of each respondent to be included in the survey. The paper presents the application of a location sampling method used for surveying EU citizens with Bulgarian migration background in Hamburg. At the beginning, arguments for the usage of location sampling are provided. After that the survey conduction is overviewed and main results are presented. Furthermore, a weighting procedure for overcoming the initial bias is sketched and main characteristics in the sample before and after weighting are compared. Arguing that an unbiased sample is achieved, estimates of the total population with Bulgarian background are produced. Finally, methodological challenges and chances of the sampling technique are discussed and it is asked to which extent a survey based on this sampling frame is an apt technique for drawing a sample in a representative way for migrant groups without an access to official population registers. 2 Survey design: the choice of location sampling There are various methods for sampling of special populations such as general procedures based on simple and multiple lists, procedures for populations that are geographically clustered and for populations defined by their activities at the locations of these activities (Sudman, Kalton 1986). Due to the lack of lists providing information about persons with Bulgarian background (numbers in Microcensus are too low, the Central Register of Foreigners and the local registration office cover only Bulgarian citizens) and no geographical clusters of Bulgarians (mainly due to their small number in the total population), there is an urgent need for looking at other methods for drawing a probability sample. In case of no sampling frame, three main techniques have been utilized to construct probability samples: time-location sampling, respondent-driven sampling or probability snowballing and capture-recapture (Marpsat, Razafindratsimab 2010). These are rather samples that can be extrapolated, i.e. that allow the production of unbiased estimators with a calculable variance (Marpsat, Razafindratsimab 2010, p. 4). The time-location sampling implies that the population of interest visits a certain number of places that are not much visited by the rest of the population, [ ] these places can be sampled after an exhaustive list has been made of them, and then a sample can be taken of the persons visiting those places (Marpsat, Razafindratsimab 2010, p. 5). This type of method has been utilized in surveys on homeless, drug users and on men having sex with other men (see overview in Marpsat, Razafindratsimab 2010, p. 5). It is argued that it is far more effective to sample and interview the respondents at the locations visited by them rather than at home (Sudman, Kalton 1986, p. 424). In the migration research location sampling has been used in several studies. In a study on level of remittances received and incidence and determinants of return migration among Japanese Brazilian households of migrants, data was collected at Nikkei community gatherings, ethnic grocery stores, sports clubs and other locations where family members of migrants are likely to congregate (location survey was referred to as intercept point survey). The authors found out that the location sampling method is likely to overrepresent persons who are more closely connected to the community and suggested to select many locations and to use reweighting in order to become more representative for the target population 1 The migrant survey is carried out in the framework of a PhD thesis on the Integration situation of EU citizens in Germany using the example of Bulgarians migrants in Hamburg. The dissertation project explores the role of EU citizenship for integration from a theoretical and empirical perspective at macro and micro level.

4 4 (McKenzie, Mistiaen 2009, p. 359). Results from the location survey were compared with those of two alternative survey methods census-based sample and snowball survey (McKenzie, Mistiaen 2009). It was demonstrated that the location survey is able to provide close results to the stratified sample based on census data if the sample is reweighted by taking into account individuals who are more likely to visit multiple locations (McKenzie, Mistiaen 2009, p. 359). The ISMU Foundation in Italy has developed and employed location sampling labelled as Centre Sampling or aggregation centre technique in order to collect data on the living situation of migrants given a high share of irregular and unregistered migrants (Blangiardo 2008). The sampling technique consists of carrying out a standardized survey with migrants at selected aggregation centres under the main assumption that every migrant visits at least one centre such as consulate, church, school, migrant organizations, entertainment centres, etc. The Centre Sampling technique has allowed to reaching migrants from a certain nationality as it was demonstrated in studies on Egyptian migrants in Milan (Baio et al. 2011) and Ghanaian, Egyptian and Ecuadorian migrants in Milan (Localmultidem 2009). An experiment with the location sampling in Germany was undertaken in 2011 with Bulgarian- and Albanian-speaking persons in Berlin (Kovacheva, Vogel 2012). 2 In the study 158 Bulgarian and Albanian speaking persons were surveyed at selected cultural, religious and commercial places, but the sample was not weighted, so that no extrapolations to the total population were undertaken. However, subgroups of the population with migration background like naturalized and unregistered persons were reached, showing the applicability of the location sampling for persons with migration background in Germany. For surveying population of Bulgarian background in Hamburg in the framework of the doctoral thesis, it is opted out for location sampling for several reasons. First, a main assumption of location studies in the destination countries is that migrants tend to cluster in certain geographic locations, and to be regular visitors at certain fixed points and community events (McKenzie, Mistiaen 2009, pp. 340, 359). It is argued that the population with Bulgarian background (in short Bulgarian migrants) in Hamburg engages in activities at migrant-oriented locations such as ethnic cultural, religious and commercial places. As these centres are less often visited by persons without Bulgarian background, respondents can be easily approached. It is, however, crucial to define a wide set of centres regularly visited by Bulgarian migrants so that each person of the target population has non-zero probability of inclusion. Second, the method is cost-efficient and allows for conduction of a small-scale survey among one group with migration background. Third, it is associated with higher likelihood to offer representative results for the target population compared to other methods such as snowball sampling (McKenzie, Mistiaen 2009, p. 342). The challenge in terms of the inclusion probability (it is higher for persons with more visits at the centres than for persons who visit rarely the centres) is well known (Kalton 2009, p. 137). Statistical weighting is used taking into account the individual probability to be sampled. For that reason, questions are asked on the frequency of visiting all of the centres of interviewing. Thus the initially biased sample can be reweighted and extrapolation of results to the total population is possible. It is worthy of naming some disadvantages of the sampling: persons have less time to participate and much shorter questionnaires have to be used (McKenzie, Mistiaen 2009, p. 342). Following Sudman and Kalton (Sudman, Kalton 1986, p. 424), three main aspects to be considered when designing survey are sketched: where to sample, when to sample and how to cope with different probabilities of sampling for different respondents. 2 The method was tested for first time in Germany in the framework of a study commissioned by the Senate administration in Berlin, which explored the labour market experiences of Bulgarian and Albanian-speaking persons.

5 5 A main prerequisite of the location survey is to identify a sufficiently wide and heterogeneous set of centres, so that diverse migrant population is covered according to age, gender, education, family status, length of stay, citizenship and migration status. Unless the ISMU location survey that target general public places, predominantly migrant-oriented centres were identified with a high likelihood of coming across persons of Bulgarian origin. 12 centres were selected for data collection: Honorary Consulate of Republic Bulgaria, Bulgarian Church "Hl. Kyrill und Hl. Methodius" Hamburg, Bulgarian school in Hamburg, Bulgarian Café Thrakia, Bulgarian ethnic shop BG Kost, German-Bulgarian Association Hamburg, Bulgarian Family and Women Club, Dance club on Bulgarian folk dances Zdravets, Central Bus Station (terminal with buses departing for and arriving from Bulgaria) and migrant counselling centres with a high proportion of Bulgarian visitors (homeless shelter, advising centre for European workers, language course for Bulgarians). It is assumed that persons of Bulgarian origin visit the centres or events organized at these centres. It is essential to include centres where also the older Bulgarian community is likely to be met. However, the Bulgarian community is relatively new one and it can be expected that many Bulgarians still keep contacts to Bulgarian-related events and organizations. It is not clear whether second generation over 15 could be efficiently reached at the selected centres. When the centres are to be visited is decided on the basis of higher probabilities of visits that may lead to larger samples. The ethnic shop and café are visited on weekends and on Friday evening. The Bulgarian kindergarten and school were visited on Saturdays as there are classes only these days. The church was visited on Sundays in the morning when it is open to visitors. At centres with regular visiting hours such as the ethnic café and ethnic shop, respondents are sampled as they arrive and enter the centre, so that every person has the probability to be included in the sample. At events organized at centres such as Christmas celebrations at Bulgarian school or Bulgarian church, respondents are sampled within the location before and after the event. The survey is designed as probability sampling, using a random selection procedure. Centres are visited at selected dates and respondents are selected at random. If it is not possible to conduct the interview immediately (e.g. as the person does not have time but is willing to participate), a meeting at another place can be arranged. What is crucial is to recruit the interviewees in the centres. The number of persons to be interviewed is defined at about 400. One centre is considered as a reference point and the statistical units are distributed to the centres according to the assumed size of the centres. At small centres smaller number of respondents is to be recruited. The distribution of interviews in the Hamburg survey is as following: commercial centres (33 per cent), cultural and religious centres (30 per cent), educational and advisory centres (22 per cent), public places (5 per cent) and other places (10 per cent). The probability of inclusion in the sample depends on the individual s frequencies of visits at the centres. As different respondents have different probabilities of inclusion, weighting adjustments are undertaken based on the visits made by each respondent (inverse weighting). A profile is given to each respondent according to the centres he or she visits. It is assumed that not all persons of Bulgarian origin visit regularly the selected centres for some reasons: they do not know about their existence or avoid contacts with persons of Bulgarian origin. This means that individuals have zero probability of inclusion that makes the estimates biased (Sudman, Kalton 1986, p. 426). Therefore, the face-to-face data collection at the locations is supplemented by online data collection via mailing lists of the Bulgarian associations. A mixed-mode survey design is preferred, aiming at better coverage of the population of interest. The personal interview, self-administrated or with interviewer, is the main mode of data collection whereas the web survey has a subordinate and only complementary role (about 10 per cent envisaged interviews).

6 6 A standardized questionnaire consisting of 8 pages (49 questions) was developed in the course of the dissertation and focuses on experiences and practices of EU citizens in the socioeconomic and legal-political area and individual attitudes towards EU citizenship. 3 It was preceded by formulation of research questions based on theoretical considerations, operationalization and formulation of questions. Thus, it seeks to illustrate a snapshot of the integration situation of Bulgarian migrants in one locality (Hamburg) and gives insights into personal attitudes and intentions that are probably less contextual, so that they can be generalized for the overall Bulgarian migrant group. Questions are asked about: knowledge about EU rights, use of these rights (political participation and naturalization) and motivations for using these rights (personal attitudes, aspirations and intentions). It provides insights into the Bulgarian migrant group in Hamburg, whose migration and integration patterns have rarely been studied so far. The participation of foreigners in surveys differs from that of the German population. The chief reasons are different accessibility, language ability for participating and the willingness to participate (Blohm, Diehl 2001). The accessibility of foreign citizens and their ability to participate may lead to under-coverage while a higher willingness to participate is associated with the group. Persons without German skills are hardly to be included in surveys with questionnaires in German (Blohm, Diehl 2001, p. 224). The willingness to participate depends on the topic of the survey and is higher in case of topics perceived subjectively as of high significance for the participants (Blohm, Diehl 2001, p. 225). Therefore, for the survey with persons with Bulgarian migrants a bi-lingual questionnaire was designed: in German and Bulgarian. From the survey research it is known that there is a relation between the mode of data collection and the way of processing of questions, respectively responses (mode effects). Therefore, the paper and online questionnaires consist of the same questions. A difficulty was the construction of a questionnaire for application in different modes of data collection: selfadministrated paper and pencil questionnaire, face-to-face interview with an interviewer and web questionnaire. The questionnaire has been designed in such a way that the interview should not exceed 40 minutes. It is not a short questionnaire so that special efforts were put into increasing the willingness to participate. The topic of the survey EU citizenship and own migration experience in Germany has been perceived as relevant and led to high response rate. 3 Data collection: pretest and field work in Hamburg The questionnaire was pretested in November 2012 with the aim at checking whether the questions are understandable. Based on the feedback from the pretest, the phrasing of some questions was revised in order to be much more understandable and neutral. Minor changes of the questionnaire were undergone in the first week of the field work that, however, did not impact on the meaning of the questions. Before starting the field work, collaboration with the selected migrant-oriented centres had to be provided. All centres were directly contacted and asked for permission for the survey conduction. Most of the centres gave permission and supported the survey by allowing access to their buildings and events. Not all of the centres planned in the survey design stage could have been visited for data collection. Due to data privacy reasons no interviews were carried out at the Bulgarian consulate and the counselling centre for mobile European workers (Beratungsstelle Arbeitnehmerfreizügigkeit). No interviews were conducted at the university structure PIASTA and at offices of Bulgarian lawyers and doctors. The lack of permission of 3 Qualitative interviews were conducted with 5 migrants and 3 experts that served for the questionnaire preparation.

7 7 certain centres shows the limitations of the location sampling technique. 27 per cent of all respondents stated that they visit often or rarely the Bulgarian consulate and events organized by it, but no interviews could have been conducted at that centre. However, a check of the cases reveals that the respondents who visit the consulate declared to visit at least one of the other centres, so that the inclusion probability is not zero. The interviews were conducted by native Bulgarian speakers: the author and two recruited interviewers of Bulgarian origin who had been trained (male and female). Bulgarian speakers were preferred in order to establish credibility and mutual trust with the respondents. Interviewers were assigned to visit each location during pre-specified blocks of time, depending on the centre. Large events such as the Christmas party of the Bulgarian kindergarten and school and the Christmas concert in the Bulgarian church were visited by two interviewers in order to sample more respondents in the short period of time. Places with longer working hours (ethnic shop and cafe) were visited by one interviewer unless there was a special event. Apart from the regular opening hours of the centres, three main events organized in the field work period were visited: Celebration of Student Party on 8 December in Café Thrakia ( ), Christmas party of the Bulgarian kindergarten and school ( ) and Christmas concert in the Bulgarian Orthodox Church ( ). Field work on the ground including conversations and comments was documented. Persons visiting the centres were approached by the interviewers in Bulgarian in order to establish trust in the initial contact. There were two selection criteria for participation: age and Bulgarian migration background. Respondents had to be at least 15 years old in order to be interviewed and to have Bulgarian migration background. In this case they were invited to take part in the survey by filling out the questionnaire. If no Bulgarian background was found out, no interview took place. Respondents were free to choose between a questionnaire in German or Bulgarian and whether to fill it out by themselves or with an interviewer. As people are often reluctant to participate in surveys, the interviewers were trained to emphasize the topic of the survey as a stimulus for participation. Interviewers were trained to convince migrants to participate by: explaining the survey aims, stressing out the anonymity, telling their personal story and giving flyers with useful information. A flyer about rights of EU citizens in the EU was handed out as an incentive (Your Europe, Your Rights, A practical guide for citizens and businesses on their rights and opportunities in the EU s single market). A lottery with 4 presents was organized as a further incentive. Respondents were asked to write their address in order to participate. In the field work the survey topic appeared to be the strongest incentive for participation. The survey has been carried out between 28 November 2012 and 15 March 2013, consisting of two consequent stages. In the first stage the selected commercial, religious and cultural centres were regularly visited and face-to-face and self-administrated interviews were conducted (December-February). In the second stage an invitation to the web survey was sent via mailing list of the participating centres (mid January-March). The invitation mail was circulated via mailing lists of the Bulgarian church (300 s), Bulgarian school (65 e- mails) and German-Bulgarian association in Hamburg (53 s). A link to the survey was also posted on the facebook site of the German-Bulgarian association in Hamburg and information was included in the PIASTA newsletter of the University of Hamburg, which has organized several events on Bulgaria and is regularly visited by Bulgarian students. Table 1 shows that there were 60 appointments in the centres of interviewing and 403 questionnaires were filled out by eligible respondents and analyzed. The most frequently visited places are the ethnic shop (16 per cent), church (13 per cent) and cafe (12 per cent). Looking at the number of visitors among the respondents, the more than the half visited the ethnic shop and almost the half visited the church (48 per cent) and the cafe (43 per cent).

8 8 Table 1 Centres of interviewing Nr. Type of centre Number of visits Number of interviews Share of all interviews Number of visitors* Share of all visitors (403) Share of all visits (1,450) 1 Consulate School Church Cafe Shop Cultural association Club of Bulgarian families and women Dance group Bulgarian doctors, lawyers et cetera Central bus station Advisory centres PIASTA Other places Online, and snowball x x x Total % 1, % Source: centre-oriented survey. *Respondents who declared to visit the centres often or rarely. Three modes of data collection were applied: self-administrated interview at the centre, faceto-face interview at the centre and self-administrated interview at home (including 52 cases of the online survey) (see Table 2). Self-administrated mode was preferred but particularly for low-educated respondents an interviewer was appointed. The vast majority was filled out by the respondents immediately in the centre while in 31 per cent of the cases the interviewer conducted personal face-to-face interview with the respondents. Table 2 Modes of data collection Modes of data collection Number Share self-administrated questionnaire in the centres face-to-face interview it the centres self-administrated at home (incl. online) Source: centre-oriented survey. The duration of interviews was between 20 and 40 minutes, on average 30 minutes. Face-toface interviews took more time than self-administrated questionnaires. If the respondent was willing to participate but did not have time in the centre, the questionnaire was handed out for home. 155 questionnaires were taken home and 89 of them were given back to the interviewers. That corresponds to a response rate of 57 per cent. This was particularly meaningful in the case of ethnic shop where visitors did not have much time to participate in 30-minutes survey and of events due to short time before and after the event. The vast majority opted for a questionnaire in the language of country of origin (92 per cent) while merely 8 per cent preferred the German questionnaire. Interviewers were trained to recommend the Bulgarian version of the questionnaire particularly to persons without excellent German knowledge as particularly questions on EU citizenship required high level of German skills. The application of bi-lingual questionnaires enabled the participation of persons without German skills who would have not participated in the survey due to lack of

9 9 German knowledge. 96 respondents declared that they read bad or not at all German, for whom it can be assumed that survey in German would not have covered them. This corresponds to 24 per cent of the sample. 4 Sample description: weighting, quality and estimates In this section the weighting procedure is described, quality of survey data is discussed and estimates of certain characteristics of the target population are presented. Weighting The initial sample is biased due to different probability of inclusion of individuals. The individual probability of inclusion depends on the number of the centres visited by the respondent and the number of individuals in the population who visit each of the centres (Baio et al. 2011, p. 453). The more units visit the centre, the lower weight it has and vice versa, and the more centres the unit visits, the lower individual weight he or she has. Therefore, the sample is weighted in order to take account of frequency of visits to each centre declared by each unit. A weight coefficient of each centre is computed and based on that, a weight coefficient is attributed to each respondent, taking into account the frequency of visits to the centres of interest. Apart from the centres visited by each individual, it is asked about the frequency of visits to them. 4 The attendance is expressed by a value between 0 and 1: 0 = if the respondent does not visit the centre 0.5 = if the respondent visits the centre rarely 1 = if the respondent visits the centre often The weights for seven centres that were selected for computation of weights are shown in Table 3. The church is considered as a baseline centre. Lower value for the centres shop, dance group and advisory centres is associated with lower weights, showing their lower relevance compared to the baseline centre. The centres school, cafe and cultural association receive higher weights, showing their higher relevance compared to the baseline centre. Table 3 Value of weights of individual centres Centre Weight of centre School 1.81 Church 1.00 Cafe 1.73 Shop 0.39 Cultural association 1.02 Dance group 0.80 Advisory centres 0.74 Source: centre-oriented survey. Based on the weights of each centre and the frequency of visits to the centres declared by each respondent, an individual weight is computed. The weight can be interpreted as the number of individuals in the target population represented by the respondent (Biemer, Christ 2009, p. 317). Each unit is assigned the corresponding weight and the sample is weighted. Thus the weights are used in order to account for differences in selection probability. The weighted 4 Question: How often do you visit the centres listed below?

10 10 sample overcomes the original bias and allows for extrapolation of sample results to the total population of interest. 5 Comparison between main socio-demographic characteristics of respondents in weighted and non-weighted sample reveals intriguing results (see Table 4). Table 4 Socio-demographic characteristics in weighted and non-weighted sample Characteristic After weighting Before weighting Impact of weighting Number Share Share Number of respondents Gender male female Age missing values Education compulsory education (8 th class) secondary education (gymnasium) secondary education (vocational school) tertiary education (university) other == Citizenship Bulgarian German other == Among them: dual citizenship (Bulgarian and German) Length of stay less than 1 year years years years more than 20 years missing values Ethnicity based on language skills only Bulgarian-speaking Turkish-speaking Romanes-speaking Ethnicity based on religion Christian-Orthodox Muslims Other religion No religion Source: centre-oriented survey. The weighting procedure had a great impact on gender, education, ethnicity and citizenship whereas age and length of stay changed rather slightly. The share of men increased from 52 to 57 per cent as opposed to the share of women that declined from 48 to 43 per cent. As to age, 5 For more information on the computation of weights, see (Baio et al. 2011).

11 11 the share of the age group rose as opposed to the share of those aged and more than 65 that declined. Non-negligible changes occurred in the educational structure particularly in the lowest and highest educational level. The share of persons with compulsory education (8 th class) almost doubled from 14 per cent to 25 per cent as opposed to the share of those with tertiary education that dropped from 50 per cent to 37 per cent. Remarkably, the share of respondents who obtained their highest educational level in Germany decreased from 37 to 29 per cent. With regard to citizenship, the share of respondents with German citizenship dropped from 18 per cent to 11 per cent and correspondingly that of dual citizens from 14 per cent to 9 per cent. 88 per cent of the respondents are Bulgarian citizens, 11 per cent German citizens and more than two third of them are dual citizens. Merely one respondent possess citizenship different from Bulgarian and German. As to length of stay, the share of respondents with short-term stay in Germany (up to 5 years) rose from 57 per cent to 66 per cent while the share of long-term stayers (more than 5 years) declined at 34 per cent. Interesting shifts in the ethnic composition occurred. There are two big minority groups in Bulgaria: ethnic Turks and ethnic Roma. The Roma are Christian Orthodox or Muslim and speak Romanes that has many different dialects (Koinova 1999, p. 150). The Turkish minority speak Turkish and lives mainly in north-eastern and south-eastern Bulgaria (Koinova 1999, p. 149). The vast majority of the Muslims are ethnic Turks, but there are also other groups like Bulgarian Muslims, Muslim Roma and Tatars (Koinova 1999, p. 155). Conclusions on the ethnic composition of the sample can be drawn based on two characteristics: language and religion that are considered a sign for belonging to ethnic minority. The share of both characteristics pointing to ethnic minorities increased as a result of weighting. The share of both Muslim respondents and Turkish-speaking respondents increased: from 16 per cent to 27 per cent for the former and from 22 per cent to 37 per cent for the latter. The share of Romanes-speaking increased from 4 per cent to 7 per cent. Correspondingly, the share of Bulgarian-speaking and Christian-Orthodox respondents declined. It has to be kept in mind that the number of Turkish-speaking respondents probably includes also some Roma and ethnic Bulgarians originating from regions with high concentration of ethnic Turks, so that the share of Turkish-speaking is considered as a maximum share of the Turkish minority. Among the Muslim respondents there are probably few Roma and Bulgarian Pomaks but there are also a large number of Turkish-speaking respondents who did not declared religion. Therefore, the share of Muslim respondents is considered as a minimum share of the Turkish minority. Accordingly, there are between 27 per cent and 37 per cent ethnic Turks in the sample. It is argued that the Romanes language is a good criterion for belonging to the Roma minority. Thus 7 per cent of the respondents in the sample are considered ethnic Roma. The results of the weighting processing give indications about the behaviour of Bulgarian migrants. After weighting the shares in the sample declined for women, young persons aged 15-44, highly educated, German citizens, long-term stayers and ethnic Bulgarians, so that they tend to visit the centres more frequently than men, persons aged 45-64, persons with low educational level, Bulgarian citizens, short-term stayers and ethnic minorities that tend to visit the centres less frequently. From methodological point of view, it came out that the former subgroups are more likely to be included in the sample which may lead to their overcoverage. It is easy to understand that women, young persons and ethnic Bulgarians seem to seek contact to the Bulgarian community more frequently. However, interestingly, the ethnic bias holds true rather for long-term migrants and naturalized persons than for newcomers and Bulgarian citizens. An explanation could be a need of well-established migrants to preserve contacts with the Bulgarian community as opposed to newcomers that seek to gain ground in the country by intensive contacts with the receiving society than the own community.

12 12 Furthermore, newcomers are probably not aware enough of the existing community centres and visit only few of them. Quality One of the criteria for sample quality is the response rate although not in all cases a higher response rate means better survey quality. The response rate is the number of complete interviews divided by the number of eligible respondents in the sample and there are various ways to calculate it (AAPOR 2011, p. 44). The calculation of response rate for a location survey is difficult due to the lack of list of eligible respondents. For that reason, the number of visitors at every visit in a centre that are considered potential eligible respondents was noted. In case of large events the number of potential eligible respondents was approximated. The response rate is calculated based on the formula of the American Association for Public Opinion Research (AAPOR) 6 that was adjusted for the location survey: Response rate = complete interviews/((complete interviews + partial interviews) +(refusals + break offs) + other) Table 5 Calculation of response rate Calculation of response rate Number complete interviews 403 partial interviews 0 refusals I (in the centres) 546 refusals II (at home) 66 break offs I (in-person) 0 break offs II (web) 72 other I, no eligible respondents (in-person) 2 other II, no eligible respondents (internet) 31 response rate 0,36 *adjusted response rate 0,42 Source: centre-oriented survey. The total response rate accounts for 36 per cent. It is argued that the actual response rate is higher as some respondents visit several centres and/ or the same centres more than once which may lead to double counting of the same persons in the number of refusals. 7 Moreover, at certain events not all visitors were asked to participate in the survey due to time constraints. Assuming that 25 per cent of the refusals are non-eligible respondents, the total response rate increases from 36 per cent to 42 per cent. With the aim at comparing response rates of different modes, response rates for each of the survey modes are calculated. Respondents who received the questionnaire in the centres but filled it out at home have the highest response rate (57 per cent), followed by respondents interviewed in the centres (35 per cent) and respondents who filled out the online questionnaire (34 per cent). The response rate of the online questionnaire is low as there are many respondents who were not eligible for participation (mainly as they do not live in Hamburg). For the in-person mode it is meaningful to calculate a response rate for all inperson interviews carried out in the centres. The response rate accounts for 46 per cent. The response rate is influenced mainly by the refusals. The refusal rate varies from centre to centre. The number of refusals was high at the central bus station and in the ethnic shop (50 per cent) and low in religious, cultural, educational and advisory centres (10-12 per cent) The number of refusals is calculated as the difference between number of visitors and conducted interviews.

13 13 Whether findings from a survey can be generalized to the total population is an important issue to discuss. The term representativeness is often used as an argument for generalization of results. It is mainly associated with the idea that the sample is representative if its structure is similar to that of the total population (von der Lippe 2011, p. 25). However, the concept is criticized as inadequate quality criterion due to a lack of a theory or a mathematical model to show how to interpret smaller or larger representativeness (von der Lippe, Kladroba 2002, p. 10). Some authors argue that the quality of a sample and an estimate should be assessed on the basis of sample error (von der Lippe, Kladroba 2002, p. 10). It is important to look at the: sample size, sample error (standard deviation of the sampling distribution), confidence intervals and variance of the characteristic of interest in the total population as an expression of homogeneity of the total population. Even though the concept of representativeness it controversial, it is worthy of comparing the weighted sample with official data from a methodological point of view. A comparison between estimates of demographic characteristics with the characteristics of the target population is one of the methods for assessing coverage (Lohr 2009, p. 103). Taking the registered population with Bulgarian citizenship as a reference point (see Table 6), a comparison of the sample shows that the shares estimated in the sample are close to the values for the population. Table 6 Bulgarian citizens in Hamburg at Number Share Total foreign population in Hamburg 241, Bulgarian citizens in Hamburg 4, Gender Male 2, Female 1, Age (18+)* , more than Source: Statistik Nord data from local population registers, own compilation. *No data is available for population less than 15, so that 18 years was taken as a reference point. The Bulgarian registered population in Hamburg accounts for 4,289 persons in It is the fourteenth country of origin for Hamburg with Turkey, Poland and Afghanistan being the three biggest sending countries. The male population dominates with 55 per cent over the female population. The vast majority of the population aged 18+ is between 18 and 45 years. Only 1 per cent is elderly people aged over 65. The gender and age distribution in the sample corresponds to a large extent to the distribution in the registered population. No data on length of stay is available at the local level. Migration is age-selective process in which persons in the first half of the working life migrate more than persons in the second half of the working life (Wagner 1989). Therefore the share of persons aged between 18 and 45 can be taken as an indicator for the relevance of newcomers. According to local registration data 80 per cent of the population is aged between 18 and 45. For comparison, the share of among the Turkish population accounts for 60 per cent and among the Polish population for 64 per cent. That points to the relevance of young persons, which migrate to Germany. No reference data on education of the Bulgarian population is available. The sample points that more than one third of the respondents have tertiary education and more than one third completed secondary education. One fourth obtained compulsory education up to 8 th class.

14 14 The high share of persons with high education can be explained by the great relevance of student migration in the Bulgarian migration pattern to Germany. No official data by ethnic group is available in the German statistics but indications about the ethnic composition of the group can be gained by data from the sending country. According to the Census carried out in 2011 the vast majority of the population in Bulgaria belongs to the ethnic Bulgarian group (84.5 per cent) and two big ethnic minorities: Turkish (8.8 per cent) and Roma (4.9 per cent). 8 Survey results show over-representation of the Turkish minority group that lies probably between 27 and 37 per cent of all respondents. An explanation could be a strong social network effect of the population with Turkish background in Germany that probably pulls Turkish-speaking people to the country. Another explanation could be overrepresentation of the ethnic group in the sample coverage. Estimates Arguing that the sample is unbiased after the weighting procedure, survey results may be extrapolated to the total target population and used for estimating certain characteristics. Two estimates are produced: of the German citizens with Bulgarian background and of the unregistered Bulgarian citizens. Based on them, the total number of the population with Bulgarian migration background in Hamburg is estimated for In the questionnaire the citizenship of respondents was captured. 11 per cent of the respondents are German citizens with Bulgarian background. Applying this percentage to the registered Bulgarian population in Hamburg, the number of German citizens with Bulgarian background is estimated at 472 persons. Furthermore, it was asked whether the respondent is registered with the local registration office or not. 46 respondents declared that they are not registered, 2 did not know and 1 refused to answer. Assuming that they are not registered, there are additionally 12 per cent unregistered Bulgarian migrants. 10 Thus the number of unregistered Bulgarians in Hamburg in 2012 is estimated at 515 persons. Adding up the three categories, registered Bulgarian citizens, German citizens with Bulgarian background and unregistered Bulgarian citizens, the total number of the population with Bulgarian background in Hamburg at is estimated at 5,275 persons. Among them there are 81 per cent registered Bulgarian citizens, 10 per cent unregistered Bulgarians and 9 per cent German citizens. The share of German citizens with Bulgarian background does not seem implausible, looking at indications from naturalization data at national level. About 10,000 Bulgarians naturalized between 1997 and Assuming that the vast majority of them remained in Germany, the ratio between naturalized persons and Bulgarian population is 1:10. Likewise, the ratio between German citizens and the total population in the estimate is 1:9. The vast majority of German citizens are dual citizens. This can be explained by the fact that most of them naturalized after 2007 since then no renouncement of the citizenship is required for EU citizens. Respondents who naturalized before 2007 lost their Bulgarian citizenship. A large precision of an estimate based on a sample requires a large sample size (Hox et al. 2009, p. 9). If the sample consists of 384 respondents, the width of 95% confidence interval is +/- 5% (Hox et al. 2009, p. 9). As the sample achieved consists of 403 respondents, it is argued that the width of 95 per cent confidence interval is +/- 5%. 8 accessed: Data on foreign population for 2011 are used. The estimate will be updated when data for 2012 are available. 10 A study conducted in Berlin revealed a similar share of 12 per cent unregistered persons. Based on police criminal statistics, it was argued that such a share is not implausible (Kovacheva, Vogel 2012, p. 30).