An Assessment of the Generalizability of Internet Surveys

Abstract

The Internet has become an increasingly popular form of data collection because it permits complex questionnaires to be administered more quickly, flexibly, and inexpensively than conventional survey methods. However, the Internet is restricted to individuals with access to computer networks. Thus, causal inferences to the general population from analyses of Internet samples necessarily rest on two untested assumptions: (a) that the decision-making processes of Internet users are similar to those used by the general population, and (b) that representative samples of Internet users can be drawn. The authors provide mixed support for these assumptions. They find that current Internet sampling techniques only permit the generation of diverse, not representative, samples. However, comparing samples drawn simultaneously using the Internet and probabilistic telephone methods, the authors demonstrate that the psychological mechanisms underlying common political decisions do not differ between Internet users and the population. They discuss the implications of these findings for future survey research.