Using an RDD-recruited Web-enabled panel to collect survey data combines the convenience of self-administered Web data collection with the statistical rigor of probability sampling.
However, Web-enabled panels generally suffer from low overall response rates due to the nonresponse at the recruitment, installation, screening, and cooperation stages. Non-response follow-up surveys (NRFUS) have been employed in past studies to adjust the population estimates. The NRFUS is conducted via CATI, which may introduce bias due to the change in data collection mode from the main study. This study examines the potential for mode bias in NRFUS using a split sample design with CATI and Web-enabled data collection.
The survey was conducted by collecting independent samples of panel members via telephone and a Web panel supported by Knowledge Networks. Follow-up on non-respondents was also conducted. The mode effect is measured for this study, and statistical methods of weighting the observations are examined--through which the mode effect could be reduced in future studies involving the Web-enabled panel. The effect of any adjustment upon the variance of the estimates is also examined.
|