487 – Challenges in the Evaluation of Biologics
Dealing with Nonresponse Using Follow-Up
Michael Hidiroglou
Statistics Canada
Victor Estevao
Statistics Canada
Non-response in surveys is usually addressed using imputation or weighting adjustments, given that no follow-up sample has been selected. These procedures are reasonable if the respondents and non-respondents are similar. If they are not, these procedures may lead to serious biases. A follow-up sample of the non-respondents will eliminate (attenuate) any such bias. However, since not all units selected in the non-response sample will respond, this leads to a three-phase sampling design. In this paper we provide weighting procedures that properly account for this, as well as the associated population and estimated variances that reflect the weighting. Also, given that a sample of the non-respondents will be followed up, we provide the allocation between the initial sample and follow-up sample size given cost constraints.