Censoring of outcomes and regressors due to survey nonresponse: Identification and estimation using weights and imputations (Q1379914)

The paper deals with ambiguity in identification of population parameters due to survey nonresponse which is practically taken into account through the use of weights, imputations or, more generally, models of nonresponse processes [see \textit{C.F. Manski}, Identification problems in the social sciences. Harvard Univ. Press (1995)]. It is argued that, avoiding untestable assumptions on the distribution of missing data, one still can establish bounds for parameters. Also the estimates using weights are potentially more (asymptotically) biased than those using imputations, whereas efforts to increase response can substantially improve identification. The finite-sampling aspects of inference play a subordinate role in these considerations. Actually, the analysis is focused on estimating certain conditional expectations involving the characteristics of population members. This problem is studied within different (including novel) schemes of data censoring, as a manifestation of nonresponse [cf. the present authors, Econometrica 63, No.2, 281-302 (1995; Zbl 0820.62030)], under fixed or variable survey organization. Illustrative empirical examples are given.