Integrating migrants into representative cross-sectional and longitudinal survey designs 2 |
|
| Chair | Professor Jürgen Schupp (SOEP/DIW Berlin ) |
| Coordinator 1 | Professor Lucinda Platt (LSE) |
| Coordinator 2 | Professor Narayan Sastry (ISR/Univ Michigan) |
The German Adult Education Survey of 2016 on behalf of the Federal Ministry of Education and Research (BMBF) and as part of the EU-AES 2016 was oversampled with regard to the target group of persons with migrant background. Thus, there are two samples drawn in a different way: a) sample of target persons with migrant background as part of the main AES-2016-survey (app. 800 persons in the age of 18 to 69 years) and b) additional sample of migrant population (n=700 aged 18 to 69 years).
Whereas subsample a) was drawn via random route due to ADM-standard as CAPI, the additional sample b) was drawn in a three-step sampling. Firstly, approximately 50.000 contacts were gathered in the method of random route due to ADM-standard. These addresses were coded by an onomastic technique in order to identify those households with a certain probability of having migrant background. The resulting sample of app. 5.000 addresses identified in this way lead to about 700 CAPI-interviews. Several measures have been taken to increase response of the target migrant group.
The oversampling of persons with migrant background shall give more information on following questions: a) Is there a positive selection in the migrant sample of the German main AES sample with regard to educational background? b) Does the AES overestimate educational participation of persons with migrant background?
A definition of the target group due to AES-instruments will be introduced. A comparison of both samples by method and content will be given.
Authors:
Frauke Bilger and Alexandra Strauß (Kantar Public Germany), Prof. Dr. Halit Öztürk and Sara Reiter (University of Muenster)
New immigrants to the US have grown steadily as a fraction of the population over the past 20 years. Despite a slowdown in immigration due to the Great Recession, new immigrants (defined as those who moved to the US in the past two decades, since 1997) now comprise approximately 6% of US families and 11% of US children (due to higher fertility among new immigrant families). Most established longitudinal surveys, including the US Panel Study of Income Dynamics (PSID), follow the same sample of families over time. The only way for new immigrants to be observed in the survey is if they marry into a sample family; hence, families comprised entirely of new immigrants have no way to enter the survey. PSID, which begun in 1968, added a sample of new immigrants once before: in 1997, a total of 511 post-1968 new immigrant families (amounting to approximately 7% of the sample at that time) were added following a major screening effort.
In 2016–2017, PSID embarked on a new effort to screen, recruit, and interview a sample of new immigrants. The goal is to add a sample of approximately 600 new post-1997 immigrant families to the survey, which would represent approximately 6% of the sample of 10,000 families. However, the costs of screening such a narrow segment of the population are extraordinarily high. Even with high response rates and oversampling areas with higher concentrations of new immigrants, we would need to screen approximately 20,000 households to achieve sample goals. Fortuitously, an opportunity to collaborate on screening operations with the Health and Retirement Study (HRS) emerged.
HRS in 2016 is screening a large number of households to select a new cohort of respondents (born between 1960 and 1971), and agreed to add PSID new immigrant screening questions to households that did not meet the HRS selection criteria. Cases that meet the PSID new immigrant screening criteria are being transferred to us for a second stage screening that establishes eligibility and collects detailed contact information. These cases will become part of the PSID sample to be interviewed for the 2017 fieldwork wave. A final crucial piece of our plan is to fill-in the donut hole created by HRS taking all cases with individuals who were born between 1960 and 1971. We will do so by creating a “multiplicity” sample from the new immigrants’ relatives (parents, siblings, and adult children) and identifying and recruiting to the sample new immigrants from among this group. Information needed to define and select the multiplicity sample will be collected in the 2017 PSID interview, and eligible new immigrant families will subsequently be invited to join PSID.
In this presentation, we will describe the results of the 2016 PSID new immigrant screening operations and discuss our plans for interviewing new immigrants in PSID in 2017, including the design and implementation of the multiplicity sampling approach. We will end by summarizing some of the major lessons we learned for other studies of new immigrants.
Minorities are often under-represented in surveys; i.e. national background is a strong predictor for survey inclusion in Switzerland, as shown by an analysis of three big Swiss surveys. In the light of relatively high percentages of foreign resident population, this represents a thread to the representativity of surveys by limiting their scope to the analysis of the majority rather than to represent the variability in the population. One cause for the under-representation, which is addressed in this paper, is the challenge coming from identifying members of minority groups. The target group maybe a small minority which will result in too small effectives for analysis even in the case of proportional inclusion in the sample. Additionally, selection from register data may be impossible in many cases due to data protection issues or the lack of visibility of the required characteristic in official databases. Network sampling methods are often used when appropriate sampling frames are lacking., i.e. in public health studies, to recruit members of hidden populations or minorities hard to access with the usual sampling strategies. Network sampling uses ties between individuals to recruit participants from within people’s contact networks, normally performing several waves or iterations until the desired sample size is reached. The LIVES Cohort Panel Study (LCPS) is, to our knowledge, the first study which uses a network sample in a large national survey. The sample was added as a subsample of the Swiss Household Panel (SHP) and will be interviewed within the future waves of the SHP. The aim of the study was to over-sample second generation immigrants born between 1988 and 1997. Second generation immigrants were defined by criteria which are not available for sampling in official registers (having been at school in Switzerland by their 10th birthday at latest; both parents immigrated to Switzerland after their 18th birthday). The study used a controlled network sample with unequal selection probabilities. The sampling strategy combined three elements in order to attain the objective: (1) A stratified random starter sample (2) a screening of the starter sample with the objective of obtaining a sample with two thirds of second generation immigrants and (3) two subsequent network sampling iterations with screening by proxy and weighted randomized selection of eligible peers. We will discuss the advantages and inconvenients, as well as practical implications of network sampling methods as a mean of recruiting survey samples. The sampling design allowed to obtain a sample which over-represents the target group and where population weights can be calculated due to known inclusion probabilities. However, the method also presents some challenges. In the first place, the dependence of the observations has to be accounted for in the analysis. Further, the choice of the starter sample is crucial and may be challenging. Finally, network sampling also has practical implications. One important aspect is the willingness of the participants to disclose their personal networks and to provide contact information. Also, the subsequent sampling iterations depend
Obtaining high quality samples of hard-to-reach groups such as migrants is challenging but vitally important in the current context as European policymakers look to measure progress with integration. The European Union Agency for Fundamental Rights (FRA), with Ipsos MORI, have recently completed the second European Union Minorities and Discrimination Survey (EU-MIDIS II), which aims to do exactly that. The survey set out to achieve a probability sample of the main immigrant and ethnic minority groups in all 28 EU Member States. Where feasible, population registers that allowed direct identification of eligible members of the population were used for sampling. Alternatively, frames that permitted estimation of target group concentrations at a local level were used for sampling in a two-stage selection process in conjunction with screening by interviewers in the field.
However, there were a number of countries where no such frames were available/accessible, or the target groups in question were so rare in the population that screening was inefficient. Here a location sampling approach was used. These countries included Austria (targeting Sub-Saharan Africans), Cyprus (Asians), Denmark (Sub-Saharan Africans), Malta (Sub-Saharan Africans), the Netherlands (two target groups: North Africans and people of Turkish descent), Poland (recent immigrants from outside the EU) and Sweden (two target groups: Sub-Saharan Africans and people of Turkish descent). In four of these countries location sampling was used in conjunction with a two-stage clustered sample with screening (Austria, Cyprus and the Netherlands) or population register (Poland).
This presentation will set this context and describe the procedures used to obtain a location sample, how the samples were weighted and dual-frame samples combined, and will present an assessment of the accuracy of results. The procedures developed for this survey were based on those elaborated in Baio et al (2011). The method involves extensive preparatory work to identify a diverse set of locations in a given area (e.g. a city) that taken together could offer good coverage of a target group by virtue of their visitation patterns. Interviewers then attend the selected locations and randomly select respondents and screen them to recruit eligible ones into the sample. The method allows estimation of selection probabilities by collecting information on the locations visited by the sample, whereby weights can be derived based on the relative importance of each location estimated from the data (detailed in Baio et al).
Sample accuracy will be assessed by comparing sample estimates with those from external sources where available. In addition, we will discuss the weighting procedures and their impact on the data. Finally, we will describe some of the challenges that presented themselves during implementation of the method and how they were overcome. These included particular challenges in finding locations/respondents when the population was very small, and covering target group populations that are less homogeneous.
Reference: Gianluca Baio, Gian Carlo Blangiardo, Marta Blangiardo. Centre Sampling Technique in Foreign Migration Surveys: A Methodological Note. Journal of Official Statistics, Vol. 27, No. 3, 2011.