Web Survey Bibliography

Title Disentangling relative mode effects for the web survey mode in the Safety Monitor
Year 2011
Access date 30.11.2012
Full text

PDF (135 kb)

Presentation

PDF (164 kb)

Abstract

Web is one of the four survey modes that is employed at Statistics Netherlands for social surveys. The most common design is a mixed-mode design where self-administered modes are offered to households first. When a household does not respond within four weeks, then interviewer-administered modes are used as follow-up survey modes. Since the web survey mode is conjectured to produce nonresponse and coverage errors, it is not used as a single-standing mode. Mode effects, i.e. differences in total survey error attributable to the survey mode, impact the comparability of surveys in time and over the main publication domains. The coverage of web is shifting gradually in time. When different subpopulations respond to different modes and answers are to some extent mode-dependent, then longitudinal statistics about such subpopulations cannot be compared directly. In order to estimate and disentangle mode effects into mode-dependent coverage, nonresponse and measurement effects, Statistics Netherlands conducted a large-scale split-ballot experiment linked to the Safety Monitor. A sample of persons was randomly assigned to the four survey modes, web, paper, telephone and face-to-face. The full sample received a face-to-face second wave in which part of the questions was repeated. The Safety Monitor was selected as the survey instrument because it is known to be sensitive to various mode-dependent measurement errors.
The analysis focusses on five research questions:

  • How do survey mode, socio-demographics and nature of the survey question interact with selection and measurement effects?
  • What role do survey design features play in selection and measurement effects?
  • Do regular nonresponse adjustment methods effectively deal with selection effects in mixed-mode surveys?
  • Can we distinguish and predict mode-dependent response styles?
  • Do standard measurement error models detect mode effects accurately?

In the paper we restrict ourselves to the first research question and discuss relative coverage, nonresponse and measurement effects for web relative to face-to-face. In section 2 we describe the experimental design. In section 3 we discuss estimators for the various mode effect components. We end with a number of general questions for discussion in section 4.

Access/Direct link

Workshop Homepage (abstract) / (presentation)

Year of publication2011
Bibliographic typeConferences, workshops, tutorials, presentations
Full text availabilityFurther details
Print