A 2-STAGE IMPUTATION PROCEDURE FOR ITEM NONRESPONSE IN SURVEYS

Regardless of the overall response rate, surveys that involve individual self-reports frequently have some degree of item nonresponse. Rather than discard these individuals from any analysis that is to be conducted and reduce the sample size, the typical procedure used by researchers is to simply substitute the mean value computed from all respondents to the question as the response for any individual not answering the question. The main problem with such a procedure is that it assumes item nonresponse is random, which often is not the case, and therefore, could produce biased results. This article proposes and empirically demonstrates a 2-stage procedure for imputing missing values. The procedure first tests whetther item nonresponse is systematic or not systematic. If nonresponse is found to be non-systematic, then a mean substitution method is acceptable. However, if item nonresponse is systematic, the second stage of the procedure imputes answers for each individual who does not answer the question using a modified least-squares procedure outlined in the article. © 1991.