Endorsement Experiment
Also known as: Endorsement question design, Endorsement experiment design, Indirect support measurement, Group-endorsement experiment
An endorsement experiment indirectly measures latent support for a sensitive or stigmatized actor by randomizing whether a policy is attributed to that actor and comparing how respondents' support for the policy shifts. Formalized statistically by Bullock, Imai, and Shapiro in 2011 to measure support for militant groups in Pakistan, the design infers favorability toward an actor that respondents would not safely disclose directly from the change in policy support it induces, typically estimated with hierarchical item-response models.
Key highlights
- Elicits support for dangerous or stigmatized actors without asking respondents to incriminate themselves, reducing nonresponse and false denial.
- Provides protection and plausible deniability that improves data quality and respondent safety in high-risk settings.
- The hierarchical item-response framework yields a continuous latent-support measure with full uncertainty quantification and covariate mapping.
- Pools information across multiple policy items, improving precision relative to a single indirect question.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use an endorsement experiment when the quantity of interest is support for an actor so sensitive, stigmatized, or dangerous that direct questions would yield refusals, evasions, or untruthful answers, and when there exist credible policies the actor could plausibly endorse. It is well suited to measuring sympathy for militant or insurgent groups, controversial political movements, or stigmatized organizations across a population. It is less appropriate when no believable policy linkage exists, when the actor's stance on the policies is ambiguous, or when sample sizes are too small to support the hierarchical estimation, since indirect designs need more data than direct questions.
Strengths & limitations
- Elicits support for dangerous or stigmatized actors without asking respondents to incriminate themselves, reducing nonresponse and false denial.
- Provides protection and plausible deniability that improves data quality and respondent safety in high-risk settings.
- The hierarchical item-response framework yields a continuous latent-support measure with full uncertainty quantification and covariate mapping.
- Pools information across multiple policy items, improving precision relative to a single indirect question.
- Requires larger samples and more sophisticated modeling than direct questions to recover the latent support signal.
- Validity hinges on the endorser being credibly linked to the policies; implausible attributions break the design.
- The measure is relative and model-dependent, expressed on a latent scale rather than as an intuitive proportion of supporters.
- Endorsement effects can conflate genuine support for the actor with reactions to the actor's competence or salience rather than warmth.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How does an endorsement experiment differ from a list experiment?
Both are indirect techniques for sensitive attitudes, but they elicit different things. A list experiment estimates the prevalence of a sensitive item by comparing counts of affirmed statements between a list with and without that item, yielding a proportion. An endorsement experiment instead measures support for a sensitive actor by observing how attributing policies to that actor shifts policy support, yielding a latent favorability estimate. Endorsement designs are well suited when naming the actor at all is unsafe, whereas list experiments target the prevalence of a specific behavior or belief.
Why use a hierarchical item-response model rather than a simple difference in means?
A simple difference in support for each endorsed versus unendorsed policy is informative, but it treats each policy separately and ignores that the same latent support parameter drives shifts across all of them. The hierarchical item-response model pools information across policies and respondents, estimates policy-specific discrimination and difficulty, and recovers a single continuous support scale with proper uncertainty. This improves precision and lets analysts relate latent support to covariates, which a per-item difference cannot do coherently.
Can an endorsement effect be negative, and what does that mean?
Yes. If attributing a policy to the actor lowers support relative to the unendorsed control, the estimated endorsement effect is negative, indicating that the population is, on average, hostile toward the actor. Heterogeneity is common and informative: the same endorsement may raise support among some subgroups and depress it among others. Care is needed in interpretation, because a drop could reflect distrust of the actor's competence or judgment rather than ideological opposition.
Sources
- 1.Bullock, W., Imai, K., & Shapiro, J. N. (2011). Statistical Analysis of Endorsement Experiments: Measuring Support for Militant Groups in Pakistan. Political Analysis, 19(4), 363–384.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 22). Endorsement Experiment. ScholarGate. https://scholargate.app/political-science/endorsement-experiment