When 'No Liars' Fails: Bounding Sensitive Answers in List Experiments

🔎 Problem Addressed

List experiments rely on two assumptions: 'no design effect' and the stronger 'no liars' assumption. The 'no liars' assumption often fails in practice. This paper relaxes that assumption and develops a method to estimate bounds on the prevalence of sensitive behaviors or attitudes under a weaker behavioral assumption about respondents’ truthfulness toward the sensitive item.

📊 Where the Method Was Applied

• A list experiment measuring anti-immigration attitudes among California residents
• A broad set of existing list experiment datasets drawn from prior applications

🔧 What the Method Does

• Provides partially identified bounds for the prevalence of a sensitive item when respondents may not always follow the 'no liars' rule
• Relies on a weaker behavioral assumption about truthfulness toward the sensitive item rather than full truth-telling

📌 Key Findings

• The width of the prevalence bounds depends jointly on:
• The prevalence of the non-sensitive items on the list
• The correlation structure among items on the list
• Bounds are generally narrower when list items belong to the same category (examples include multiple social groups or organizations, related corporate activities, or several considerations that politicians weigh)
• The method clarifies when the full 'no liars' assumption is necessary to precisely identify prevalence and when weaker assumptions still yield useful, robust estimates

⚖️ Why It Matters

The approach helps researchers and practitioners diagnose how fragile list-experiment estimates are to violations of the 'no liars' assumption and provides a practical way to report prevalence estimates that remain valid even when respondents sometimes misreport the sensitive item.