Abstract

Sexual orientation is a sensitive subject and; consequently, in a survey, nontrivial alternatives of item nonresponse arise. Although handling these item nonresponses is particularly important when the missing information is not missing at random, little is known about the sources of nonrandomness. Using data drawn from the 2013–2018 National Health Interview Survey and the multinomial logit model, this study results revealed that participants’ socio-demographic characteristics are systematically correlated with the occurrences of missing values in the “don’t know” and “refused to answer” response alternatives of the sexual orientation question even when there is controlling for other variables of interest. The results suggest a need for greater attention when researchers conduct an analysis using a complete data set for which pairwise or listwise deletion of the item nonresponse is applied.