| Nonparametric models and nonresponse adjustments in surveys
| Jean D. Opsomer
Ph.D., Iowa State University
Tuesday, February 6, 2007
C101 Plant Sciences
Nonresponse is a common problem in surveys, which can lead to biased estimators and erroneous inference if it is not accounted for in the estimation procedures. Because the nonresponse mechanism is typically unknown, nonparametric specification of this mechanism is attractive because of its generality and robustness. We will address two issues related to nonresponse. First, we evaluate the robustness of the commonly used "weighting cell estimator" to misspecification of the cells, by assuming a nonparametric response probability mechanism. We show that the weighting cell estimator remains consistent even under cell misspecification. Second, we develop a new response-weighted estimator that relies on a nonparametrically estimated response probability function. The theoretical properties of the new estimator are derived, and a replication-based variance estimator is proposed.