Surveys, Social justice, Migration and Face-to-face |
|
| Session Organisers | Jorg Blasius Susanne Vogl |
| Time | Friday 16 July, 15:00 - 16:30 |
Dr Mariel Leonard (German Center for Integration and Migration Research (DeZIM)) - Presenting Author
Dr Sabrina Mayer (German Center for Integration and Migration Research (DeZIM) & University of Duisburg-Essen)
Dr Jörg Dollmann (German Center for Integration and Migration Research (DeZIM) & Mannheim Center for European Social Research (MZES), University of Mannheim)
Sampling procedures are a factor in determining a survey’s data quality. Whereas many easy-to-use procedures exist for the general population, approaches for sampling small, hidden and/or hard-to-reach populations are more difficult. Many individuals with a migration background or ethnic minorities are part of these challenging populations, which hinders our possibilities to study their every-day life, experiences, and attitudes. Relying on name-based procedures (onomastic classification) for pre-qualifying person addresses is by most considered the gold standard at the moment. Such a procedure is not always feasible as it is cost-intensive and relies on a large number of personal addresses in the pre-qualifying sample in order to identify a sufficient number of members of small groups that can serve as the gross sample. Furthermore, it has been shown that using onomastic classifications increases the risk of systematically missing well-integrated individuals. Currently, several other approaches have been tried such as WebRDS or using social media ads but so far, they have not been tested systematically against each other.
Our research aims to close this research gap. In 2021, the German Centre for Integration and Migration Research (DeZIM) launched the DeZIM.panel, a non-commercial online access panel that is supposed to be representative for several major groups of people of immigrant-origin for which we employ onomastic classification. For this research project, we focus on the group of Portuguese migrants and their children (i.e., 1st and 2nd generation). Since the 1960s, Portuguese migrants have entered Germany in three waves: first as "guest workers" in the 1960s-1970s, then following the fall of the Berlin Wall, and finally a third wave that is ongoing since the 2010s. While the population of Portuguese Germans is relatively small (under 200,000 individuals) and geographically concentrated in the German states of Nord Rhein-Westphalia and Baden-Württemberg, the population varies greatly with respect to age, length-of-residence in Germany, reason for migrating, degree of integration, and other important characteristics.
In addition to the name-based approach, we test the efficacy and representativeness of three other common methods used to sample ethnic minorities and other hard-to-reach groups: (web) respondent-driven sampling, social media-based convenience sampling, and center (time/location) sampling, to determine which performs better. We evaluate these methods based on their cost-effectiveness in providing a representative sample of Portuguese Germans participating for the DeZIM.panel. We compare representativeness of the samples to the German Microcensus and cost-effectiveness to a probability-proportionate-to-size sample of Germans of all ethnicities. Our paper will provide an overview of our implementation of each method, including the strengths and weakness of each; our evaluation criteria; and preliminary results.
Dr Blazej Palat (CDSP) - Presenting Author
Mr Valentin Brunel (CDSP)
Face-to-face recruitment operations for surveys or panels in social sciences are often very similar. The first contact step consists almost always of sending an information letter to the whole sample. During ELIPSS panel’s refreshment, the CDSP has worked with La Poste and CDA towards a renewal of classic recruitment processes. For the first phase of the operation, information letters and leaflets were handed-in directly to sampled people when possible, resulting in roughly a half of the sample receiving information about the panel in a more noticeable way before being asked to join it by La Poste a week later. After this first phase of recruitment, a survey institute proceeded to a second exploitation of the sample, which led to our final response rate.
In order to examine the impact of information-letter distribution patterns (handed in personally by the postmen vs. left in the mailbox) in the first phase of the recruitment process on the recruitment efficacy by the end of that phase, we fitted a logistic regression model with the recruitment success as a binary response variable. Control variables were derived from socio-demographic information originating from the population sample file. We found that handing in information letters personally facilitated the recruitment process (, dwelling type, surface, occupant status, age, time for which he/she occupied it, presence, and age of children in the dwelling being controlled for). This model did not include either observations where the postman was unable to find the requested individual, or those where information letters, and therefore participation in the panel were directly refused by that individual in face-to-face contact with the postman. Thus, we also fitted an additional multinomial logistic regression model including cases of direct refusal, and excluding cases where the letters were left in the mailbox. Our aim was to calculate the odds of postmens’ failure to find the requested individual, or seeing the information letter refused by him/her directly in face-to-face contact rather than being able to hand it in. This model included the same independent variables as those used as control variables in the logistic regression model. We found that similar variables to those that predicted recruitment success also predicted failure to find the requested individuals, and that direct refusals were mainly related to those individuals’ age.
This study seems to point out a positive correlation between having information on the panel handed in personally and taking part in it. This conclusion theoretically satisfies our expectations regarding the importance of strengthening the bond between the panel and its respondents.
What’s more, effects from the first phase of the recruitment process could be measured on two other important steps in the ELIPSS panel recruitment: second phase of recruitment and first online survey. We suppose that, although the relation between being handed-in a letter and accepting to participate in phase 2 or answering the first online survey can weaken, the effect of a friendly postman’s gesture will persist.
Ms Sandra Gilgen (University of Bern & LMU) - Presenting Author
Questions of distributive justice affect so many areas of our lives, spanning the private as well as the political, societal spheres. Understanding what people think is fair and why is thus highly relevant for the understanding of many outcomes of human interaction in many contexts, from the home to the workplace to the development of policies and institutions. While the subject of distributive justice has received much attention from many fields, including economics, sociology, psychology, and social anthropology, generalisable mechanism-based findings have been scarce. A reason is that depending on their individual situations, the contexts that have shaped their preferences and beliefs, people apply different principles of justice, the most important of them being: equality, merit and need. While these individual-level differences, such as in regard to class and gender as well as effects of the context, such as in the form of cross-cultural comparisons have received considerable attention, the fact that our perceptions of justice are very much shaped by the situation has not. To be able to measure effects on attitudes to distributive justice on all three levels: individual, contextual and situational, the distributional survey experiment was developed as a new tool. In this survey experimental design, respondents are asked to distribute a specified amount of money among three people described in vignettes. This approach combines the possibilities of distribution tasks in laboratory settings with the interdependency and visual presentation of a choice experiment and the convenient metric outcomes of factorial surveys. An advantage of the method is that the situation can easily be adapted, to include allocation tasks in different settings, such as among family members, friends, students applying for scholarships and at the workplace. Additionally, the method captures the nature of the problem of questions of (re)distribution, which is that it involves trade-offs between competing principles of justice but also that at its core, outcomes are interdependent. Giving more to some leaves less over for the others, which is captured in the design of the distributional survey experiment. Most importantly however, applications of the method show that in comparison to effects of individual respondent characteristics and the context, the situation is a key component in any justice evaluation.
Dr Hayk Gyuzalyan (Highgate Consultancy) - Presenting Author
Quantitative research among ethnic minorities has a few methodological challenges. One of the challenges is not knowing the prevalence of such groups in the general population. It creates problems for generalisations from the sample to the universe, makes it difficult to correctly estimate the selection probabilities of those included in the sample, and makes it difficult to ensure the possibility of participation for all members of the population. Several approaches are used to esitmate the prevalence, each having their own
We are developing an approach which combines some advantages of other methods. We use a directory of names from ethnic minority community organisations in a country, match the names in the directory with the list from a publicly available directory (White or Yellow pages), estimate the proportion of matches and estimate the number of people in the community organisation who are not registered in the public available directory. Calculating these proportions will result in a fairly accurate estimate of the prevalence of ethnic minorities in the country.
The benefit of the method is that there is no need to run a full match, and no need to use a very large number of names, to ensure that all common names are included. The approach can use only a small number of names (ie those starting with two or three letters of the alphabet), and by estimating the proportion of matching in the public available directory, and the corresponding proportion of matching in the community organisation directory, it is possible to estimate the prevalence of ethnic minority in the country.
The approach is based on one assumption, that the names of a given ethnic minority are mostly unique to this ethnic minority, which is correct for many ethnic minorities, but not all.
Dr Sebastian Wenz (GESIS) - Presenting Author
Usually, survey researchers turn to traditional regression models that focus on conditional means—e.g., linear regression or hierarchical linear models (HLMs)—when assessing differences between immigrants and natives in metric outcomes, such as test scores in educational assessments, wages, or various measures of health. These models estimate the conditional mean function, E(y|x)= xβ, and, thus, estimate differences between means of immigrants and natives. However, such models ignore that differences in means between immigrants and natives—or any other two or more sociodemographic groups, for that matter—are not necessarily the same as differences in other points (e.g., the 10th percentile, the median, or the 90th percentile) of the outcome’s distribution.
In my contribution I show how quantile regression (Koenker & Bassett 1978, Koenker 2005) provides a convenient way of assessing differences between immigrants and natives beyond the mean by estimating the conditional quantile function, Q_tau(y│x)=xβ_tau, and, thus, the difference between quantiles tau, 0
I conclude by a brief discussion of the most important limitations and misconceptions around quantile regression: In contrast to popular belief, the traditional conditional quantile regression estimator (Koenker & Bassett 1978, Koenker 2005)—without further assumptions—neither allows inference about particular observations at the quantile under study, nor does it provide estimates that can readily be interpreted as changes in quantiles or differences at quantiles of the unconditional distribution of the outcome. Also, the Quantile Regression Model (QRM) neither requires much less assumptions than the Linear Regression Model (LRM), nor is it any better suited for or in any way superior to the LRM when modeling nonlinear relations.