Mixing Protected and Direct Survey Answers Shows When Sensitive Techniques Are Needed

🔎 Why current survey tricks fall short

Sensitive survey techniques (SSTs) are widely used to measure taboo or stigmatized behaviors, but their use often produces highly variable prevalence estimates. Existing SST strategies also leave a crucial question unanswered: is the SST necessary in the first place for a given item?

🧭 What the new approach does: combine crosswise and direct answers

This article introduces a questioning strategy and a statistical framework that jointly analyzes responses from an SST—the crosswise model—together with direct (unprotected) answers about the same sensitive behavior. The joint framework:

• Tests whether an SST is actually required to obtain unbiased responses for a given behavior
• Produces an efficient estimate of the behavior's prevalence by leveraging both protected and direct responses
• In an extended form, estimates how individual characteristics relate to the likelihood of engaging in the sensitive behavior

📊 How the method works in practice

• The crosswise model supplies privacy-protected data while direct responses provide unprotected but potentially informative reports.
• Joint analysis reconciles these two sources, reducing reliance on either one alone and improving efficiency of prevalence estimates.
• The framework also yields estimates of covariate effects (e.g., demographic predictors) in an efficient manner when the extended model is applied.

🧪 Demonstration with Costa Rica: gender and corruption proclivities

• The approach is illustrated through an application examining gender differences in proclivities toward corruption in Costa Rica.
• That demonstration shows how combining protected and direct responses can clarify whether SSTs are necessary for detecting prevalence and for estimating individual-level correlates.

✅ Why this matters

This integrated strategy gives survey designers and analysts a practical way to decide when to deploy SSTs, to obtain more reliable prevalence estimates, and to efficiently estimate relationships between respondent characteristics and sensitive behaviors—reducing the variability and uncertainty that currently plague SST-based estimates.