USING SURVEY PARTICIPANTS TO ESTIMATE THE IMPACT OF NONPARTICIPATION

We consider models that underlie two proposals to estimate nonparticip ation bias. The first model posits a ''continuum of resistance,'' plac ing people who were interviewed during the first contact on one end of the continuum and nonparticipants on the other. The second model assu mes that there are different classes of nonparticipants and that simil ar classes can be found among participants; it then uses groups of par ticipants thought to be like nonparticipants to estimate the character istics of non-participants. We examine the justification for these mod els of the relationship between participants and nonparticipants and c onsider how well proposed methods based on these models describe nonpa rticipants and the impact of nonparticipation on survey estimates. The case we analyze is estimats of means of child support awards and paym ents in Wisconsin. We find that neither model is successful and that t he versions of the methods we use do not detect the true extent of non participation error in estimates based on the unadjusted sample mean. This failure occurs both for an external measure that is not contamina ted with response errors and for self-reports. But response errors, wh ich are not considered in the models we have found in the literature, substantially worsen matters.