Attrition in the Austrian Generations and Gender Survey: Is there a bias by fertility-relevant aspects?

Isabella Buber-Ennser

doi:10.4054/DemRes.2014.31.16

Volume 31 - Article 16 | Pages 459–496

Attrition in the Austrian Generations and Gender Survey: Is there a bias by fertility-relevant aspects?

By Isabella Buber-Ennser

Date received:

3 Feb 2014

Date published:

22 Aug 2014

Word count:

8333

Keywords:

attrition, Austria, dropouts, Generations and Gender Survey (GGS), longitudinal data, response rates

DOI:

10.4054/DemRes.2014.31.16

Replicable Material:

demographic-research.31-16 (zip file, 118 kB)
readme.31-16 (text file, 345 Byte)

Download PDF
Submit a Response Letter

Abstract

Background: In longitudinal research the loss of sample members between waves is a possible source of bias. It is therefore crucial to analyse attrition.

Objective: This paper analyses attrition in a longitudinal study on family and fertility, by distinguishing between attrition due to non-contact and attrition due to non-cooperation.

Methods: Based on the first two waves of the Austrian Generations and Gender Survey, the two components of attrition are studied separately by using bivariate as well as multivariate methods. Moreover, overall dropout – the combination of both components – is analysed. Apart from various socio-economic characteristics and data collection information, the study focuses on fertility-relevant variables such as fecundity, fertility intentions, sexual orientation, and traditional attitudes.

Results: Fecundity, fertility intentions, and homosexual relationships are associated with higher attrition due to non-cooperation in bivariate analyses, but have no explanatory power in the multivariate model. Pregnancy and traditional attitudes towards marriage are associated with significantly lower attrition due to non-cooperation in the multivariate context. Overall dropout is significantly lower only among persons with traditional attitudes towards marriage, although small in size and statistical significance. Moreover, various individual and regional characteristics are significantly associated with dropout, with differences between attrition due to non-contact and attrition due to non-cooperation.

Conclusions: Detailed insights into attrition are not only important when using longitudinal data and interpreting results, but also for the design of future data collections. The Austrian GGS panel has a relatively low dropout (22%) and is affected by a small bias towards familyoriented persons as well as less-educated respondents and persons with migration backgrounds, but the data can be used without concern about selectivity.