Self-reported behavioral data are vulnerable to reporting biases, including social desirability bias, which can distort prevalence estimates of sensitive behaviors. List randomization is an indirect response technique intended to reduce such bias by masking individual responses within aggregated counts. The authors evaluated whether list randomization changes prevalence estimates for sexual-health behaviors among women of reproductive age living with HIV in Lilongwe, Malawi.
Analyses used data from the Family Planning and Antiretroviral Therapy (FP-ART) cohort. The sample comprised women of reproductive age living with HIV who participated in the study; the publication provides cohort and visit-level analyses but does not report additional demographic breakdowns in the abstract beyond socioeconomic stratification by roof type.
Participants were randomized to respond to five blocks of true/false statements using one of two modes: direct response (answering each statement individually) or list response (reporting the number of true statements in a block). Each block included three nonsensitive statements and one sensitive statement. The sensitive statements concerned either condom use or HIV disclosure.
List randomization was implemented by embedding the sensitive item among nonsensitive items; respondents assigned to the list arm reported only the total count of true items per block, preserving the privacy of responses to any specific sensitive item.
For each sensitive statement, the authors calculated the prevalence difference (PD) comparing estimates obtained under list response versus direct response. Analyses were performed overall and stratified by socioeconomic status, where socioeconomic status was approximated by roof type (rudimentary roof indicating lower status vs metal roof indicating higher status). Confidence intervals were reported for key PD estimates.
Across the set of sensitive statements, list randomization produced negligible differences in estimated prevalence for four of the items; the abstract indicates little meaningful change for most questions.
One notable exception concerned the item on always using a condom. At visits using list response, self-reported prevalence of always using a condom was 53.1%, compared with 34.7% at direct response visits. The resulting PD was 18.5% with a 95% confidence interval of 6.2% to 30.7%, indicating a statistically notable increase in the estimate under list randomization for this positive behavior.
When stratified by socioeconomic status (roof type), the elevated estimate for always using a condom under list response was attenuated among participants with higher socioeconomic status (metal roof), suggesting effect modification by socioeconomic indicators.
For the other sensitive statements related to condom use or HIV disclosure, the PDs were reported as negligible in the abstract; the publication’s figures present prevalence comparisons and stratified results, but the abstract does not list numeric PDs for each item beyond the condom-use example.
In this setting, list randomization did not substantially change prevalence estimates for most sensitive sexual-health items among women with HIV. The method produced a higher estimated prevalence for always using a condom—an unexpected direction for an indirect technique intended to reduce underreporting of stigmatized or socially undesirable behaviors. The attenuation of this effect among participants of higher socioeconomic status suggests contextual influences on how list randomization affects reporting.
The authors interpret these results to mean that list randomization may not consistently improve measurement of sensitive behaviors across items and populations. The single item with a meaningful change was a positive behavior (always using condoms) rather than a stigmatized or disfavored behavior, highlighting that indirect methods can affect reporting in unanticipated ways.
The article includes figures comparing prevalence estimates from list response versus direct response visits. Figure 1 shows prevalence differences for sexual-health behaviors using cohort data from the FP-ART study. Figure 2 displays the same comparisons stratified by roof type as an indicator of socioeconomic status, with separate panels for rudimentary roof (lower socioeconomic status) and metal roof (higher socioeconomic status). The abstract reports that stratification by roof type demonstrated attenuation of the condom-use difference among higher socioeconomic groups. Numeric PDs for most items are not provided in the abstract; the full text and figures contain additional graphical detail.
The trial found that list randomization did not meaningfully alter prevalence estimates for most sensitive sexual-health statements in this sample of women with HIV, with one exception: an unexpectedly higher estimate for always using condoms under list response. The authors conclude that examining list randomization in other settings and populations is warranted to determine its utility and generalizability for measuring sensitive behaviors.
The abstract provides key design elements, the primary notable numeric result (condom-use PD and 95% CI), and the high-level conclusion. Additional numeric results for the other sensitive items, detailed participant characteristics, and full methodological specifications (for example, exact wording of blocks, randomization procedures, or sample size per arm) are available in the full text (PMCID: PMC13160244) but are not enumerated in the abstract. The authors reported funding sources and stated no conflicts of interest.