Coverage error in internet surveys: can fixed phones fix it?
Instituto Universitário de Lisboa (ISCTE-IUL)
Widespread internet access and its broad use for a variety of purposes have been contributing to the adoption of the internet for survey work. As internet surveys are more cost efficient than other modes and make larger samples affordable, it is not surprising that many survey research companies are making the internet their mode of choice in surveys (e.g. Einhart 2003; Potoglou & Kanaroglou 2008; De Leeuw 2010). Either as online panels or cross-sectional surveys, the use of the internet for survey purposes is widespread in several domains (e.g. Comley 2007; Couper 2007; Göritz 2007; Tortora 2009).
Although internet-based surveys are on the rise, the use of the internet to conduct inferential surveys poses some concerns. Coverage error, sampling error and non-response error are major threats to inference in internet-based surveys. From a survey methodology point of view, internet-based surveys have potentially bigger problems of coverage error than traditional survey modes. In national population surveys the coverage error is generally lower for face-to-face and postal modes in which samples are drawn from lists of addresses, electoral rolls or population registers (although the availability of such lists varies considerably between countries). In telephone surveys the level of coverage error depends mostly upon the proportion of households in the population without fixed telephones. By definition, online panels or any cross-sectional internet-based surveys do not cover non-internet household members, and thus are not representative of this population’s subgroup. This is analysed in further detail in the next section.