PROBING “DONT KNOW” ANSWERS: EFFECTS ON SURVEY ESTIMATES AND VARIABLE RELATIONSHIPS

Abstract

Don't Know responses to survey questions are ambiguous because the same words are used by respondents to mean different things—ignorance, indecision, or uncertainty about the meaning of the question asked. In order to clarify the meaning of such answers, survey interviewers are frequently trained to probe Don't Know answers at least once before the answer is considered final. We argue that unconditional probing of Don't Know answers may not be a desirable practice, particularly as regards knowledge items. Large unintended effects on responses to four knowledge items resulted when two groups of interviewers, who administered the same survey questions, probed Don't Know responses at different rates. Strong evidence is provided in the article that the probing of Don't Know answers encouraged guesswork on the part of uninformed respondents, giving rise to significant distributional differences and differences in means across half samples for the affected variables. However, relationships between variables appear to be largely unaffected by probing effects. In order to formulate valid probing guidelines for interviewer training, further research is needed to establish whether the findings for knowledge questions generalize to factual questions.