Context Effects in Attitude Surveys - Effects on Remote Items and Impact on Predictive Validity

The authors present the results from parallel experiments in two surveys about privacy attitudes. Concerned that the order of the questions might affect the answers, they systematically varied the order of some of the key questions in the questionnaire. In both surveys, four of five question order experiments produced significant effects on responses to the items involved, but the question order variables did not affect responses to attitude items that came later in the questionnaire or relations between the attitude items whose order the authors varied and background characteristics of the respondents. They were also able to determine for most respondents whether their household had mailed back its census questionnaire. Question order did not have a consistent impact on the correlations between the survey responses and actual census returns. These results suggest that question order effects may be common with conceptually related items but that their impact is generally local, affecting answers to the items themselves but not answers to later questions, correlations with respondent background characteristics, or relations to subsequent behaviors.