581.5
What's the Impact of Coverage Error to the Study of Social Inequalities?

Friday, July 18, 2014: 11:30 AM
Room: 416
Oral Presentation
Emanuela SALA , Dipartimento di Sociologia e Ricerca Sociale, Università di Milano Bicocca, Italy
Roberto LILLINI , Dipartimento di Sociologia e Ricerca Sociale, Università di Milano Bicocca, Italy
Laura FUMAGALLI , University of Essex, United Kingdom
In some European countries, including Italy, survey organizations use the Directory of Landline Phone Numbers as sampling frame to survey the general population and study social inequalities (Häder and al. 2012). Under certain conditions, the use of this sampling frame may be problematic.

In Italy, with the only exceptions of Callegaro’s work (2004 and 2008), there is very little interest in this topic. Little is known, for example, on the extent to which coverage error occurs and its impact on survey estimates, although in 2011 only 50% of households are included in the sampling frame (e. g., the directory of landline phone numbers). We believe a clear understanding of the nature of coverage error is urgently needed (households may be excluded from the sampling frame for different reasons) to develop appropriate strategies to tackle the coverage problem.

The overall aim of the paper is to evaluate the impact of coverage error on the accuracy of the survey estimates for a set of key socio-economic variables and, should evidence for bias arise, to discuss possible strategies to correct for it.

We first describe changes over time in the pattern of coverage error by focusing on three groups of Italian households  (household included in the sampling frame, households  excluded from the sampling frame as they do not have a telephone and households excluded from the sampling frame as they do have a telephone but the telephone number is not listed). We then explore whether there are any differences in these groups in their socio-economic characteristics, including indicators of general trust and, for a selection of survey items (including income, health and social class), investigate the impact of coverage error on bias. We use the 1997-2012 Multiscopo survey (“Aspetti della vita quotidiana”) run by the Italian Statistical Institute (n=20.000 households).