How Do Self-Selected Internet Samples Differ from Representative Samples?

Using opt-in Internet samples for political data is far less expensive than most forms of traditional probability sampling. In some analyses of opt-in samples, researchers have found that opt-in sample estimators provide good approximations of electoral results.  In others, the differences have been both highly significant and substantively important. These conflicting results have led some to posit that opt-in samples should work effectively for certain types of inference. To assess this possibility, we explore the relationship between pairs of large samples with data collected simultaneously using both RDD telephone and non-probability Internet methodologies.  We assess the extent to which the samples provide for similar inferences under various conditions.
Work in conjunction with Jon Krosnick, Stanford University
This entry was posted in Projects. Bookmark the permalink.

Leave a Reply