Relations Between Variables and Trends Over Time in Rdd Telephone and Nonprobability Sample Internet Surveys

Abstract

Survey researchers today can choose between relatively higher-cost probability sample telephone surveys and lower-cost surveys of nonprobability samples of potential respondents who complete questionnaires via the internet. Previous studies generally indicated that the former yield more accurate distributions of variables, but little work to date has explored the impact of mode and sampling on associations between variables and trends over time. The current study did so using parallel surveys conducted in 2010 focused on opinions, events, behavioral intentions, and behaviors involving that year’s Decennial Census. A few comparisons indicated that the two data streams yielded similar results, but the two methods frequently yielded different results, often strikingly so, and the results yielded by the probability samples seem likely to be the more accurate ones.