Imputation for nonmonotone nonresponse in the survey of industrial research and development

Nonresponse in longitudinal studies often occurs in a nonmonotone pattern. In the Survey of I ndustrial Research and Development (SIRD), it is reasonable to assume that the nonresponse mechanism is past-value-dependent in the sense that the response propensity of a study variable at time point t depends on response status and observed or missing values of the same variable at time points prior to t. Since this nonresponse is nonignorable, the parametric likelihood approach is sensitive to the specification of parametric models on both the joint distribution of variables at different time points and the nonresponse mechanism. The n onmonotone nonresponse also limits the applic ation of inve rse propensity weighting methods. By discarding all ob served data fro m a subject afte r its first missing value, one can create a datas et with a monotone ignorable nonresponse and then appl y established methods for ign orable nonresponse. However, di scarding observed data i s not desira ble and it may result in ineffici ent estimators when many observed data are dis carded. We propose to i mpute nonrespondents through regression under imputation models carefully created under the pa st-valuedependent nonresponse mechanism. This method does not require any parametric model on the joi nt distribution of the variables across time points or the nonresponse mechanism. Performance of the e stimated means based on the p roposed imputation method is investigated through some simulation studies and empirical analysis of the SIRD data.