The rise of large internet consumer panels for survey research raises the question of whether samples of respondents drawn from these panels are representative of the underlying population. To assess this question we model the attrition/selection process for one major consumer panel, maintained by Knowledge Networks, Inc (KN). Starting from KN’s over 525,000 random-digitdialed (RDD) panel-recruitment telephone contact attempts, and ending with a sample of respondents to an actual online survey, we span all junctures at which systematic selection could occur. Our analysis begins by matching addresses or telephone exchanges to the appropriate census tract for the initial half-million residential telephone numbers in the RDD contact-attempt pool. This permits us to use a set of fifteen orthogonal factors based on census tract characteristics, plus county voting percentages in the 2000 Presidential election, to look for neighborhood characteristics that influence whether an initial RDD contact attempt eventually results in a usable response to a specific survey with a sample size of 2,911. We then examine how non-random selection into the estimating sample affects respondents’ answers to one specific survey question about the proper role of government in environmental, health and safety regulation. Using two distinct approaches, we do find evidence of modest sample selectivity. However, we find that these selection effects are not statistically significant in explaining respondents’ attitudes about the proper role of government in society.
Supplementary materials: Referee comments on POQ submission; Referee comments on SSCORE submission