Selection Bias of Internet Panel Surveys

Abstract

Purpose. To investigate the selection bias of an Internet panel survey organized by a commercial company. Methods. A descriptive study was conducted. The authors compared the characteristics of the Internet panel survey with a national paper-based survey and with national governmental statistics in Japan. Results. The participants in the Internet panel survey were composed of more women, were older, and resided in large cities. Regardless of age and sex, the prevalence of highly educated people in the Internet panel survey was higher than in the paper-based survey and the national statistics. In men, the prevalence of heavy drinkers among the 30- to 49-year-old population and of habitual smokers among the 20- to 49-year-old population in the Internet panel survey was lower than what was found in the national statistics. Conclusions. The estimated characteristics of commercial Internet panel surveys were quite different from the national statistical data. In a commercial Internet panel survey, selection bias should not be underestimated.