ESRA 2017 Programme
|ESRA Conference App|
Friday 21st July, 11:00 - 12:30 Room: Q2 AUD1 CGD
Representing the population: Improving European sampling practices 2
|Chair||Dr Annette Scherpenzeel (SHARE – Survey of Health, Ageing and Retirement in Europe )|
|Coordinator 1||Mrs Johanna Bristle (SHARE – Survey of Health, Ageing and Retirement in Europe)|
|Coordinator 2||Dr Stefan Zins (ESS-GESIS)|
Session DetailsObtaining good probability samples is a key challenge for European cross-national studies in order to represent the population. Four large cross-national surveys: the European Social Survey (ESS), the Survey of Health, Ageing, and Retirement in Europe (SHARE), the Gender and Generations Program (GGP) and the European Values Study (EVS) have put their efforts together to improve and harmonize the sampling practices in their studies, This is done within the consortium “Synergies for Europe´s research infrastructures in the social sciences” (SERISS). The session aims to make the results of this effort available to other survey practitioners and exchange experiences with them. It is subdivided into three themes:
1. Mapping the use of population registers as sampling frames in Europe
The availability of central population registers that can be used as sampling frames varies a lot across countries, as do the regulations about who can or can’t access the registers and what information can be obtained from them. We welcome presentations addressing the problem of obtaining good quality probability samples of the European population or national populations, the quality of alternative methods of sampling such as random route procedures, or other experiences with the use of other registers as sampling frames.
2. Including special subpopulations in surveys of the European population
The SERISS working group addresses the issue of the representation of the institutionalized population in European surveys. We invite papers addressing this topic but also papers describing sampling practices and fieldwork measures for other specific subpopulations such as refugees.
3. Using administrative data to correct nonresponse bias
Many population registers used for sampling carry additional information, such as age, gender and household composition of the registered person. Moreover, surveys can use small-area contextual data on characteristics such as neighbourhood population density, crime rates or poverty to enrich their response data. We invite researchers to submit papers which make use of administrative, contextual or other auxiliary data in Europe for the analysis of nonresponse bias or for nonresponse adjustments.
Depending on the number of submitted abstracts we might split up the session into three parts.
Paper Details1. The non-coverage of the institutional population living in retirement and nursing homes: Assessing the peril of coverage bias in a cross-national survey
Mr Jan-Lucas Schanze (GESIS – Leibniz Institute for the Social Sciences)
Dr Stefan Zins (GESIS – Leibniz Institute for the Social Sciences)
In average, about 1.3 percent of the entire European population lives in retirement and nursing homes, prisons or refugee hostels. The institutional population is usually excluded in most social surveys. Survey researchers justify the non-coverage with practical concerns and the assumed higher costs of data collection. Following Tourangeau’s scheme of hard-to-survey populations, the institutional population can be classified as hard-to-sample, hard-to-contact and hard-to-interview in comparison with the majority of private population. Our paper focuses on the relevance of people living in retirement and nursing homes for surveys. This subgroup adds up to 40 percent of the entire institutional population. Furthermore, within the European population aged 85 years or older 13 percent live in an institution; far more than in every other age cohort.
The paper is motivated by the overarching question whether surveys of the general population need additional efforts to extent the coverage due to the distinctiveness of the institutional population. Residents living in retirement and nursing homes differ significantly from their non-institutionalized counterparts in socio-demographic, socio-economic, and medical characteristics, as the 2011 European censuses and some survey research in European countries and the USA prove. Those differences could limit the ability to do inference on the whole population, if estimates are based on sample data that excludes the institutional population. Taking the deviations in demographic, economic, and geographic variables as a starting point, we conduct a semisynthetic simulation study. The simulation is based on two assumptions. The first assumption is that the population living in retirement and nursing homes deviates significantly from the respective age cohorts living in private households. They differ mainly in the distribution of gender, functional and cognitive impairments, health status, civil status, and social networks. The second assumption is that the institutional population itself is rather homogenous and depicts a lower degree of variance than the general population. On the basis of these two assumptions, we use the Monte-Carlo simulation study to detect a possible bias of sample estimates if the institutional population is systematically underrepresented or excluded from the target population. The results also allow us to assess the possibilities and limitations of using survey weights to compensate for frame imperfection caused by the exclusion of the institutional population.
2. Weighting procedures using administrative data for countering unit nonresponse bias of EU-SILC in Austria
Mr Thomas Glaser (Statistics Austria)
EU-SILC in Austria is a sample survey on income and living conditions of persons living in private households carried out on a yearly basis and using an integrated rotational design. This means that every year about one fourth of the sample is replaced by a new rotational group. EU-SILC is the main source of indicators on household income and living conditions such as the at-risk-of-poverty rate (AROP) defined as the rate of persons living in private households with an equivalised disposable income below 60% of the median.
Since participation in the EU-SILC survey is voluntary unit nonresponse inevitably happens and is most common in the first year of the survey. If response patterns differ for characteristics that are highly correlated with the main EU-SILC indicators such as AROP unit nonresponse bias will occur.
From 2012 onwards EU-SILC in Austria has been using income register data from administrative sources that are linked to the sample on micro-level as a source for collecting information about most components of the household income. Since this information is also available for the sampling frame it can be used for unit nonresponse analysis and the computation of weights. One of the most important objectives of the weighting procedure of EU-SILC is to counter potential unit nonresponse bias related to the estimators of the main indicators.
The present work compares procedures for nonresponse weighting based on inverse response probabilities or calibration. The probability of taking part in the survey can be estimated by various methods, for instance logistic regressions or pattern recognition techniques such as artificial neural networks. These methods rely heavily on the quality of register data available for respondents and nonrespondents. Calibration methods adjust the weights to external marginal distributions that also rely on administrative sources.
Finally, a Monte Carlo simulation is carried out to assess unit nonresponse bias by comparing estimators based on different weighting procedures.