ACSPRI Conferences, RC33 Eighth International Conference on Social Science Methodology

Font Size:  Small  Medium  Large

A screening or overlapping dual frame approach for telephone surveys in Europe

Femke De Keulenaer, Ahu Alanya

Building: Law Building
Room: Breakout 7 - Law Building, Room 028
Date: 2012-07-10 11:00 AM – 12:30 PM
Last modified: 2012-06-05


Experience with dual frame telephone surveys in the context of the Flash Eurobarometer has shown that, in many EU countries, a significantly different demographic mix of the adult population can be interviewed when sampling from a mobile phone frame, compared to when sampling from a landline frame. For example, in a Flash EB survey conducted in Portugal, 25% of respondents in the mobile phone survey were under age 30; this compared to just 10% in the landline sample and 22% in the population as a whole (according to population register data). This means that combining mobile phone interviews with landline interviews should lead to unweighted samples that more closely match general population parameters.

Data and methods
In this study, we compare alternative designs to integrate mobile and landline samples, weighted only to correct for estimated size of the telephone strata, in terms of which design produces the most representative sample of the general population. Flash Eurobarometer data will be used for these simulations (data from several European countries, five waves/surveys in each country).

Design 1: mobile-only (screening) design – dual users are sampled entirely from the landline frame
Design 2: overlapping design with a compositing factor λ=0.75 (larger weight for landline frame dual users)
Design 3: overlapping design with an average or multiplicity estimator (λ=0.50)
Design 4: overlapping design with a compositing factor λ=0.25 (larger weight for mobile frame dual users)
Design 5: landline-only (screening) design – dual users are sampled entirely from the mobile frame

Brick and his colleagues (2011) pointed out that many survey organisations have used a compositing factor λ=0.50 without considering the effect on bias (for example, this approach is biased in the presence of nonresponse errors). Brick and his colleagues also suggested choosing the compositing factor to eliminate the bias of the average estimator using information on the proportion of mobile-mainly and landline-mainly individuals in the population. Because such telephone usage control totals are not available for the countries included in our study, we explore modifying the estimates by using different compositing factors.