List Experiment (Item Count Technique)

Description

Control respondents count how many of several neutral items apply to them; treatment respondents get the same list plus a sensitive item; the difference estimates prevalence.

In a list experiment, respondents report only how many items on a list apply, never which. The treatment list includes the sensitive item; the difference in mean counts between groups estimates its prevalence. Because individual answers are hidden, social desirability bias is reduced. Design rules include choosing control items to avoid floor and ceiling effects (including negatively correlated items), and testing the 'no design effect' assumption. Multivariate methods (list package) allow regression on covariates. Unlike randomised response, there is no device to explain, which helps online.

Attributes

Loss type
Unmigrated
Effort
Days

Steps

  1. Choose 3-4 control items, mixing high and low prevalence and negatively correlated items to avoid ceiling effects.
  2. Randomly assign respondents to control or treatment (control list plus sensitive item).
  3. Ask only for the count of items that apply.
  4. Estimate prevalence as the difference in mean counts with standard errors.
  5. Test design effects and floor or ceiling issues; use multivariate list models for covariates.
  6. Compare with direct questions in a separate group to estimate social desirability bias.

Sources

  • Blair & Imai, Statistical analysis of list experiments (2012), Political Analysis
  • Miller, A new survey technique for studying deviant behavior (1984), PhD thesis, George Washington University

Also known as

  • Item count technique
  • Unmatched count technique
  • List experiment

Categories

  • Analytics & Behaviour
  • Survey & Psychometrics