A cross-sectional epidemiological investigation conducted across urban and rural districts of Punjab, Pakistan, demonstrated that multi-stage stratified paired response frameworks uncover substantial underreporting in both COVID-19 infection rates and vaccine hesitancy. Conventional direct-questioning surveys often fail to capture accurate public health metrics because social stigma, fear of disclosure, and nonresponse bias lead individuals to conceal sensitive health statuses. By implementing advanced randomized response techniques that mask individual replies through mathematical scrambling while preserving aggregate statistical properties, the research team demonstrated that actual disease transmission and reluctance toward immunization were markedly higher than indicated by conventional self-reported data.
The methodological foundation of this research addresses a long-standing challenge in public health surveillance: obtaining truthful measurements of stigmatized behaviors and health conditions without compromising respondent confidentiality. Building upon developments in randomized response theory by earlier researchers—including unbiased additive and optional scrambling models, variance coefficients for population estimation, and weighted measures combining efficiency and privacy—the investigators formulated multi-stage stratified paired response frameworks, designated as MSPRF-I and MSPRF-II. These models extend classical single-stage designs by incorporating dual scrambling factors within stratified sampling architectures, ensuring unbiased estimation even when adapting to complex population structures and unequal selection probabilities.
To evaluate the empirical performance of these statistical frameworks, the researchers executed a field survey between June and August 2021 targeting 1,200 adult residents aged 18 years and older across four major districts of Punjab: Faisalabad, Lahore, Multan, and Rawalpindi. The sampling strategy employed stratified simple random sampling, establishing an equal allocation of 600 participants in urban areas and 600 participants in rural areas to account for regional socioeconomic heterogeneity and balance estimator precision. Trained enumerators administered anonymous structured questionnaires in local languages, collecting demographic information alongside sensitive inquiries regarding personal COVID-19 illness history, underreporting behaviors, and vaccine hesitancy.
The operational mechanics of MSPRF-I rely on a two-stage randomized response protocol wherein each respondent provides a pair of responses to protect confidentiality. In the first response, an assigned proportion of respondents scrambles their answer, while remaining participants proceed through a second stage where two independent randomization devices generate continuous scrambling variables drawn from exponential distributions. These scrambling variables, denoted as Y and Z, are parameterized by the stratum-specific sensitivity level and the assigned scrambling fraction. In the first generated answer, the ratio of these variables is added to the participant’s genuine response, whereas in the second answer, the ratio is subtracted, creating a dual-response pair that eliminates social desirability bias without inflating data collection expenses.
The advanced MSPRF-II framework expands this procedure into a three-stage randomized response design by integrating an explicit truthful-response component alongside controlled dual scrambling. In this configuration, a predefined fraction of respondents is instructed to report their true value directly, while the remaining proportion follows the paired scrambling sequence. The mathematical formulation utilizes exponential random distributions parameterized by the assigned true-response proportion, the scrambling fraction, and the stratum sensitivity parameter. This structural modification provides a second layer of analytical adjustment, enabling the framework to separate genuine non-sensitive reporting from concealed replies while maintaining rigorous privacy protections for every survey participant.
Empirical application of these frameworks revealed pronounced discrepancies between self-reported infection histories and randomized response estimates across both geographic strata. Under the two-stage MSPRF-I model, the estimated COVID-19 infection rate reached 16.8 percent with a margin of error of 1.2 percent in urban centers and 20.8 percent with a margin of error of 1.4 percent in rural areas, compared with directly reported figures of only 12.1 percent and 13.6 percent, respectively. The three-stage MSPRF-II model produced closely aligned estimates of 16.8 percent with a margin of error of 0.8 percent among urban residents and 19.2 percent with a margin of error of 1.3 percent among rural residents, contrasting with self-reported rates of 12.6 percent and 14.6 percent.
Across the entire sampled population of the four districts, the aggregated estimates confirmed a substantial hidden disease burden and widespread reluctance toward vaccination that traditional questioning failed to register. Overall reported COVID-19 prevalence stood at 13.4 percent, whereas MSPRF-I and MSPRF-II yielded comprehensive estimates of 16.6 percent with a margin of error of 1.5 percent and 16.5 percent with a margin of error of 1.3 percent, respectively. The measurement gap was even more pronounced regarding vaccine hesitancy: direct questioning recorded a hesitancy rate of 24.3 percent, whereas the MSPRF-I and MSPRF-II frameworks revealed true underlying hesitancy levels of 33.8 percent with a margin of error of 1.6 percent and 29.5 percent with a margin of error of 1.4 percent.
The study also identified noteworthy regional and behavioral variations in disclosure patterns and disease transmission dynamics across Punjab. Estimated sensitivity parameters demonstrated that urban participants exhibited a higher sensitivity level of 0.62 compared to 0.48 among rural respondents, indicating that stigma and privacy concerns regarding infection and vaccination disclosure varied meaningfully by geographic and cultural setting. District-level evaluations over consecutive epidemiological weeks further illustrated wide dispersion in daily reported cases and substantial divergence in local sensitivity levels across Faisalabad, Lahore, Multan, and Rawalpindi, corroborating the necessity of stratified sampling rather than uniform regional assumptions.
Comparative evaluation against classical survey methodologies showed that both multi-stage paired response models achieved substantial statistical efficiency gains, demonstrating percent relative efficiency improvements ranging from 290 percent to 399 percent, and exceeding conventional randomized response procedures by up to 450 percent. Between the two proposed models, MSPRF-II exhibited slightly superior adjustment capacity and higher precision for mean parameter estimation, primarily attributable to the mathematical stability introduced by its truthful-response component. Both frameworks consistently outperformed direct questioning, establishing that mathematically bounded randomization can reliably correct for systematic nonresponse and deliberate misreporting in high-stakes public health environments.
The authors noted specific methodological limitations, emphasizing that the investigation was geographically restricted to selected districts within Punjab and that the cross-sectional survey design captures observational associations rather than temporal causal relationships. Although the dual-scrambling mechanisms and anonymous administration effectively mitigated reporting inaccuracies, future iterations of privacy-preserving frameworks could incorporate optional response pathways that allow respondents to choose freely between direct and scrambled disclosure, potentially enhancing survey cooperation without diminishing estimator efficiency.
Beyond immediate respiratory manifestations, severe COVID-19 infections documented during the pandemic caused serious organ damage, elevated mortality, and lingering long-term health complications, including profound fatigue, cognitive decline, sleep disturbances, and sensory impairments such as the loss of taste. These persistent clinical burdens placed tremendous strain on regional healthcare infrastructures, intensifying the operational necessity for trustworthy epidemiological surveillance. When public health agencies rely on flawed survey measurements, the resulting systemic underestimations weaken disease monitoring networks and impair the planning of targeted medical interventions. To capture these sensitive dynamics accurately, the field questionnaires collected comprehensive background metrics alongside infection histories, recording respondent age, sex, formal educational attainment, total household size, and monthly family income, while also evaluating local healthcare service utilization.
To maintain rigorous representative sampling across the study population, the researchers implemented replacement protocols to address survey nonresponse without introducing attrition bias. Whenever a selected individual was unavailable, enumerators substituted an eligible resident drawn from the identical geographic stratum. The operational execution of the survey relied on trained enumerators who delivered structured oral guidance and clarified procedures in local languages. While completing a double-response protocol required each participant to answer sensitive inquiries twice, this structured repetition was deliberately engineered to remain practical for real-world fieldwork. Clear preliminary instructions, explicit guarantees of total anonymity, and personal enumerator support minimized cognitive burden and prevented participant refusals, enabling the gathering of sensitive epidemiological metrics without increasing administrative expenditures.
Methodologically, the survey architecture was situated across three interconnected literature domains: theoretical developments in randomized response estimation for sensitive quantitative variables, survey methodology assessing privacy and nonresponse in health research, and applied public health investigations tracking vaccine hesitancy throughout Pakistan. Although earlier scrambling formulations concentrated heavily on equal probability sampling designs, the frameworks demonstrated adaptability to unequal probability selection schemes through the integration of sampling weights. This mathematical flexibility allows the estimators to maintain unbiased accuracy across diverse population structures. Under the stratified simple random sampling with replacement design, the partitioning of participants reduced within-stratum variance while capturing between-stratum variation. In situations where individual participants perceived a specific inquiry as non-sensitive, the protocol permitted them to report their genuine figures directly.
The empirical comparisons further illustrated that distribution patterns of daily reported infections varied widely across epidemiological weeks and individual districts, reinforcing the observed spatial heterogeneity. Graphical evaluations confirmed that the three-stage framework yielded slightly higher case estimates than the two-stage model, indicating a stronger adjustment capacity for unobserved infections and reporting hesitation. Because public health policies rely on reliable metrics to direct scarce resources, correcting for social desirability bias remains vital for effective intervention planning. In accordance with established research ethics and data protection protocols designed to preserve participant privacy, the underlying primary survey datasets are not accessible in public repositories, though identified study data can be released by the corresponding author upon reasonable academic request.
Subject of Research: Privacy-preserving randomized response survey frameworks for estimating sensitive public health outcomes
Article Title: Enhancing Estimation Precision in Sensitive Public Health Data: Implications for Evidence-Based Decision-Making
Article References: Honghe, L., Shah, F., Shahid, M. I., & Ai, K. (2026). Enhancing Estimation Precision in Sensitive Public Health Data: Implications for Evidence-Based Decision-Making. Journal of Epidemiology and Global Health, 16(1), Article 119. https://doi.org/10.1007/s44197-026-00627-7
Image Credits: AI Generated
DOI: 10.1007/s44197-026-00627-7
Keywords: randomized response technique, MSPRF, COVID-19 underreporting, vaccine hesitancy, public health surveillance, scrambling mechanisms, stratified sampling, Punjab, Pakistan, dual-response protocol, estimator precision, social desirability bias, urban-rural disparities, exponential distribution, truthful-response component, nonresponse bias, epidemiological surveys
