Empirical conclusions on national minority representation in Swiss
surveys
Driven by these motives, the eight members of the interdisciplinary working
group have coordinated their efforts over the last two years to analyse systematically
how survey non-response relates to ethno-national affiliations, across three major
nationwide surveys: the Swiss Household Panel (SHP), the Swiss Labour Force
Survey (SLFS), and the European Social Survey in Switzerland (ESS). Among other
criteria, these surveys have been selected because of their overall methodological
5
rigor and high standards. Our goal was not to highlight any particular research
project but, on the contrary, to document and start to explain minority bias as a
phenomenon that is pervasive enough to be easily discernable even within survey
research at its best. Another reason motivating our choice is that these surveys
already have implemented or experimented with practices to deal with non-coverage
or non-response, either in general (notably in the ESS) or by targeting foreigner
populations more specifically (in the SLFS).
All together, the two resulting empirical papers propose answers to nine
different research questions, all related to the representation of national minorities in
Swiss social surveys. The first set of questions (1 to 3), which were studied by Lipps et
al. (2011), address the overall issue of who is being excluded systematically, and allow
delineation of the categories and subcategories along which minority bias operates in
Swiss surveys. The second set of research questions (4 to 9, investigated by Laganà et
al. (2011)) focus on how national minorities are either excluded from or included into
surveys, and consider survey practices as explanatory factors. Let us summarise here
the corresponding answers, suggested by the findings from both papers.
Question 1: Are Swiss social surveys generally biased against national
minorities?
Yes. Self-declared national background is a very strong predictor of survey
inclusion, across all three analysed surveys. This does not mean that all foreign
nationalities are underrepresented in Swiss surveys. There are actually substantial
variations across communities. Minority bias is extreme for nationals from the
former Yugoslavia, Albania, Turkey, as well as from outside Europe (hereafter, we
will refer to these groups together as “non-Western-Europeans”), There is no
substantial bias however against nationals from neighbouring countries (Germany,
France, Italy, Austria, and Liechtenstein). Furthermore, minority bias cannot be
reduced to a class effect: even when controlling for relevant social and economic
factors, there is still a significant net bias due to national affiliations.
Question 2: Are there subcategories within national minorities that are
especially concerned?
Yes. Minority bias is strongest among the least educated. This is true in
particular among non-Western-Europeans, for which a particularly large social bias
within communities added to the national bias between communities.
Question 3: Does possible underrepresentation in cross-sections increase in
panel surveys through attrition?
Yes. Overall, attrition bias follows similar patterns as cross-sectional minority
bias. As a cumulative consequence, minority bias becomes even stronger in
longitudinal studies.
Question 4: Do additional efforts to establish contact or convert reluctant
respondents result in less minority bias?
No. From our analyses, more efforts to reach and recruit respondents using the
same survey routines lead to including more respondents of the same type. Longer
contact chains and attempts to convert reluctant respondents result in (even) more
minority bias, rather than less.
Question 5: Are more experienced interviewers more capable of reducing
minority bias?
No. In the current Swiss survey landscape, interviewer learning processes and
incentive structures seem to produce a cumulative advantage in favour of
respondents from the national majority. More experience appears to help
interviewers to recruit (even) more majority respondents and, in all likelihood, to
develop economically rewarding strategies to focus their efforts on “easy”
respondents. It therefore results in more rather than less minority bias.
Question 6: Do common weighting procedures result in statistical estimates
free of minority bias?
No. In Switzerland, a common procedure consists in adjusting survey data for
the cumulative share of all foreigners, merged into a single statistical category. This
results in the overrepresentation of minorities from close and economically
prosperous European sending countries, while the remaining minority communities
are still largely underrepresented. As a plausible consequence, weighted statistical
estimates remain largely conservative – and difficult to interpret – indicators
regarding the situation of vulnerable populations.
Question 7: Does the correction of bias between national categories, by way
of stratified sampling, also affect bias within national categories (either positively
or negatively)?
Probably not, but it might depend on the overall survey context. The larger
part of the evidence suggests that sampling with national strata is neutral with regard
to bias within national minority groups. But in combination with more survey
languages, it might even have positive effects on the representation of socially
disadvantaged groups within national categories. We still need more evidence on
possible desirable spill-over effects, but it is safe to dismiss counter-productive side
effects.
Question 8: Do additional survey languages help to recruit members from
minority groups, in particular among the socially disadvantaged?
Yes, but only to a limited extent. It seems that as long as first contacts are still
conducted in national languages, this might remain a critical obstacle to enhancing
the representation of minority communities overall, and of the socially disadvantaged
within these communities, in a more consequential way.
Question 9: Do more survey languages help to keep (socially disadvantaged)
members from minority groups in the sample of longitudinal studies?
Yes. Once minority respondents have been included in a panel study, they are
as likely as majority respondents to remain in, provided that they can be interviewed
in their own language. This is true for socially disadvantaged as well as for other
minority members.
