The Conflict Data Gap: Partial identification of official statistics under conflict-induced non-observation

Conflict and insecurity can prevent statistical agencies from observing parts of their target populations. When an accessible-community estimate is interpreted as describing the full population, inference requires assumptions about outcomes in inaccessible communities. If the same conflict process affects accessibility and the outcome, non-observation may be non-ignorable. We therefore frame the problem through partial identification rather than assuming that imputation alone resolves it. We define the Conflict Data Gap (CDG) as the width of the identification region under assumptions the analyst is willing to defend. For binary outcomes, the assumption-free CDG equals the noncoverage rate; outcome-specific ranges or directional restrictions can make it depend on both outcome structure and coverage. A Monte Carlo study with a genuine MNAR data-generating process illustrates the framework. In this design, the maintained bounds contain the true value, and regions with effectively equal coverage but different observed outcome levels have different directional gaps. In the same design, multiple imputation under ignorability yields a narrow but incorrect interval. We propose that official statistics materially affected by conflict-induced non-observation should report the CDG profile alongside the accessible-community estimate, making the resulting identification uncertainty explicit.