Structure of "Don't Know" Responses in an Information Security Attitude Survey: Implications for Multivariate Analysis of 12 Items on Surveillance, Control, Legal Regulation, and State Intervention

Abstract

This study examined the structure of “Don’t Know” (DK) responses and their treatment in multivariate analysis using 12 items on surveillance, control, legal regulation, and state intervention from a web-based information security attitude survey (n = 1,200) conducted from June 30 to July 2, 2026. A total of 702 respondents (58.5%) gave no DK responses, substantially exceeding the 25.15% expected under an item-independence model that preserved the observed DK rate for each item. The sample variance of respondent-level DK counts was 4.704, compared with 1.112 under the independence model. Mean phi coefficients between binary DK indicators were higher within questionnaire blocks, and all pairwise associations within Q15 remained positive after excluding the 87 respondents who selected DK for all five Q15 items. Respondents who selected “Neither/Neutral” (N) at least once in Q16–Q18 were more likely to select DK in Q14–Q15 than respondents who never selected N (OR = 1.688, 95% CI: 1.332–2.138); the association remained in a sensitivity analysis (OR = 2.558, 95% CI: 1.929–3.394). In an MCA treating all categories, including DK, as active, the 12 DK categories jointly contributed 72.45% to Dim1, all appeared on the same side of that dimension, and the absolute correlation between respondents’ Dim1 scores and their number of DK responses was |r| = .934. To examine the structure among substantive response categories, a specific MCA was additionally conducted in which only DK categories were treated as junk categories and excluded from axis construction. These results suggest the usefulness of distinguishing, according to the analytic objective, between analyses of DK-response structure itself and analyses that separate DK from axis construction in order to examine relations among substantive response categories.