Sanctions selection bias

Sanctions selection bias is the methodological argument, associated with Daniel Drezner, that imposed sanctions are a biased sample of economic coercion: the most effective threats succeed at the threat stage and never appear in the datasets, so studies of imposed sanctions systematically understate the instrument's coercive value. The argument reframes the sanctions effectiveness debate from "do sanctions work?" to "which sanctions do we ever get to observe?"

Mechanism

Coercion begins with a threat. If the target believes the sender will impose serious costs and expects to concede eventually, the rational move is to concede early, before costs accumulate; the episode ends quietly and no sanction is recorded. Sanctions are actually imposed in the residual cases where the threat failed to persuade: where the target discounts the sender's resolve, values the disputed policy highly, or expects to withstand or evade the pressure. The observed universe is therefore adversely selected toward the hardest cases. Drezner develops the logic in The Sanctions Paradox and tests it in "The Hidden Hand of Economic Coercion," finding that coercion episodes resolved at the threat stage generate significantly higher concession rates than imposed-sanctions episodes. The mechanism parallels selection problems elsewhere in conflict research: wars, strikes and lawsuits that actually occur are the disputes bargaining failed to settle, and imposed sanctions are the coercion attempts whose threats failed to persuade.

Implications

Three implications can be tested. First, headline failure rates from imposed-only datasets describe a selected sample rather than every coercive threat. Second, threat-stage concessions or deterrence may be omitted from those datasets, but unobserved non-action is not automatically a sanctions success. Third, senders may obtain more concessions in disputes where the target expects lower future conflict, while political incentives often produce measures against resistant rivals. For planning, the lesson attaches to positioning in the Economic Kill Chain (EKC): record threats, demands and target responses before execution, then test whether an observed concession was caused by the threat.

Contestation and limits

The extent of the bias is contested. Threat-stage episodes are hard to observe and code, so the correction risks replacing one selected sample with another. Some senders issue threats for domestic audiences with no intention of following through, which can inflate apparent threat-stage success. Measures imposed for signalling create a different outcome problem because concession may never have been the objective. Selection effects can therefore qualify imposed-only estimates without overturning them or establishing the scale of omitted success.

The correction is a research design, not a presumption that every unobserved threat worked. A dataset must define the episode, identify when a threat became observable, code the demand and outcome consistently, and preserve cases issued for domestic audiences or signalling. Threat-stage concession, deterrence of an action never attempted, third-party signalling and target compliance are different outcomes. The selection claim qualifies imposed-only estimates; it does not establish the counterfactual cause of any concession.

See also

Sanctions effectiveness debate · Economic coercion · Economic deterrence · Sanctions as signalling · Credibility and resolve · Economic Kill Chain (EKC) · Economic statecraft

Sources

• Cambridge University Press, Daniel W. Drezner, *The Sanctions Paradox*.
• Daniel W. Drezner, "The Hidden Hand of Economic Coercion", *International Organization* 57, no. 3 (2003).
• T. Clifton Morgan, Navin Bapat and Yoshiharu Kobayashi, "Threat and Imposition of Economic Sanctions 1945-2005", *Conflict Management and Peace Science* 31, no. 5 (2014).
• Peterson Institute for International Economics, *Economic Sanctions Reconsidered*, 3rd ed..
• Robert A. Pape, "Why Economic Sanctions Do Not Work", *International Security* 22, no. 2 (1997).