Privacy, Truth-Telling, and Data Quality: Experimental Evidence on Sensitive Questions Measurement in Surveys

The intervention is the mode used to administer the self-administered sections of the individual questionnaire (the mental health section and the sensitive block). Households assigned to the control arm administer these sections via text-only computer-assisted self-interviewing (CASI): the interviewer hands the device to the selected respondent, who reads the questions on screen and inputs responses directly. Households assigned to the treatment arm administer the same sections via audio computer-assisted self-interviewing (ACASI): the selected respondent uses individual headphones to listen to the questions and response options through audio and inputs responses on the device. Question wording, response options, and skip patterns are identical across arms. The sensitive block covers items on gender identity, sexual orientation, discrimination, violence, and self-harm behaviors, and also includes placebo items. Prior to the first self-administered section, all respondents complete a brief practice item to familiarize them with the assigned mode. Interviewers follow a privacy protocol during the self-administered sections, ensuring the screen is not visible to third parties and maintaining appropriate spacing.

2026-06-17

2026-08-06

(i) Disclosure of sensitive items in the self-administered block, measured as standardized family-level disclosure indices for four item families: (a) gender identity and sexual orientation, (b) discrimination, (c) violence, and (d) self-harm behaviors. (ii) Item nonresponse rates in the sensitive block. (iii) Response quality indicators for the sensitive block, derived from survey paradata and questionnaire responses.

For each sensitive item, disclosure is defined as a binary indicator equal to 1 if the respondent provides an affirmative or non-default response, 0 otherwise. The primary disclosure outcomes are family-level indices. For each family, a family-level disclosure rate is computed as an indicator equal to 1 if the respondent discloses at least one item within the family, and this rate is standardized to mean 0 and standard deviation 1 in the control arm. In addition, a covariance-weighted index is constructed for each family. Individual items are reported for descriptive and exploratory purposes. Item nonresponse is defined, for each item, as an indicator equal to 1 if the response is "Don't know," "Prefer not to answer," or missing, aggregated at the family and full-block level as the share of items unanswered. Paradata-based quality indicators are constructed from response timing and behavior captured by Survey Solutions and may include response times, item nonresponse, and completion patterns.

Survey duration (total and by module), breakoff rates (partial completion of the sensitive block), and the difference in mental health responses between the interviewer-administered (CAPI) and self-administered modes, compared across the CASI and ACASI arms.

Survey duration is measured in minutes from paradata timestamps, computed overall and separately for the sensitive block. Breakoff is defined as a respondent starting but not completing the self-administered sensitive block. The mental health items are administered twice to each respondent (once in the interviewer-administered mode and once in the assigned self-administered mode) and the difference between the two scores is computed for each respondent.

This is a household-level randomized controlled trial conducted in urban areas of the Metropolitan Region of Chile. A target of 1,260 households is randomly assigned in a 1:1 ratio to either text-only computer-assisted self-interviewing (CASI, control, 630 households) or audio computer-assisted self-interviewing (ACASI, treatment, 630 households) for the sensitive block of the individual questionnaire. Within each household, one eligible member aged 14 or older is randomly selected to complete the individual questionnaire, which combines an interviewer-administered (CAPI) component and the assigned self-administered (CASI or ACASI) component. Independently of the mode assignment, households are randomized 1:1 to one of two orderings of these two components: CAPI-first/self-second or self-first/CAPI-second. Mental health items (PHQ-4) are duplicated across the CAPI and self-administered components; because each respondent is assigned to only one self-administered mode, this allows the gap between interviewer- and self-administered measurement to be compared across the CASI and ACASI arms, with the order randomization allowing identification of this contrast net of order or fatigue effects. The household sample is selected through a stratified multistage probability design with five geographic strata corresponding to zones of Greater Santiago (Centro, Norte, Oriente, Poniente, Sur).

Not available

Both the mode assignment (CASI vs. ACASI) and the order assignment (CAPI-first vs. self-first) are determined by the Survey Solutions (SUSO) electronic questionnaire. The assigned mode and order are revealed to the interviewer by the device only after the household module has been completed and the eligible respondent has been randomly selected by the software. The interviewer has no role in, and no discretion over, the mode or order assigned to each household.

Household level. The primary intervention (CASI vs. ACASI) and the secondary order randomization (CAPI-first vs. self-first) are both assigned at the household level in a 1:1 ratio, independently of each other. Within each household, one eligible member aged 14 or older is randomly selected by the software to complete the individual questionnaire.

No

1260 households

1,260 individuals (one randomly selected eligible member aged 14 or older per household)

630 households CASI (control), 630 households ACASI (treatment). Independently, 630 households CAPI-first ordering and 630 households self-administered-first ordering. The crossing of the two randomizations yields four implementation arms of approximately 315 households each: (i) CASI / CAPI-first, (ii) CASI / self-first, (iii) ACASI / CAPI-first, (iv) ACASI / self-first.

Sample size was determined by operational and budgetary constraints, with a target of 1,260 households (630 per mode arm). We report minimum detectable effects (MDEs) assuming 80% power, a two-sided test at the 5% level, and 1:1 allocation across mode arms. Mode is randomized at the household level with one individual per household, so the unit of analysis matches the unit of randomization and no clustering adjustment is required.  The primary disclosure outcomes are family-level indices. For the binary "any disclosure" family indicator, the MDE ranges from about 4.7 pp at a baseline prevalence of 10% to 7.9 pp at 50%; because this indicator is satisfied by disclosure on any single item, its baseline prevalence is mechanically higher than that of the individual items, placing the relevant comparisons in the better-powered range. For reference, recent work documents ACASI disclosure gains of 4–7 pp for lifetime intimate partner violence (Peterman et al., 2025) and 5–18 pp for IPV indices in rural Liberia and Malawi (Park et al., 2025).