Lethal violence, not ransom, dominates the reported record. Killing appears in roughly a third of reports and release in a comparable share, while an explicit ransom demand appears in about one in six. Any framing that treats Nigerian kidnapping primarily as a ransom economy understates the lethal dimension that the media record actually describes.

The phenomenon is geographically concentrated but not confined. The North-West (Kaduna, Katsina, Zamfara) and the North-East (Borno) carry the heaviest reported burden, and these are also the areas of sparsest security presence. The South is not absent: it accounts for a substantial minority of reports, with a different profile weighted toward shorter, ransom-oriented abductions.

These figures are media-reported prevalence, not a census of true incidence. They describe what Nigerian news outlets reported, shaped by where journalists are present and which cases they judge newsworthy. High-profile mass abductions are over-represented; low-profile rural incidents are almost certainly under-counted. Read every figure on this site with that caveat, and cross-validate against ground-truth sources where decisions depend on it.

Behaviour Sequence Analysis (BSA) treats each kidnapping as an ordered chain of observable behaviours and asks which transitions occur more often than chance. The transition matrix underlying this analysis is drawn from a human-coded corpus of incidents, not from the raw media stream: journalism reports outcomes first (the inverted pyramid), so sequences read directly from article text would invert the real order of events. The validated matrix preserves the true progression.

Two resolution pathways dominate. The stronger is lethal, running from the initial seizure and movement of the victim, through captivity, to the killing of the victim. The second is economic, running from a ransom demand, through negotiation and payment, to release. Ransom is a coherent and internally consistent pathway, but it is not the organising logic of the phenomenon: across the full corpus an explicit ransom demand appears in a minority of reports, while a death appears in roughly a third.

This is the central analytical point the expanded corpus makes possible. Because the collection captures all kidnap events rather than only those mentioning ransom, the ransom and non-ransom pathways can be compared directly. The largest single category is not ransom-and-release but abduction that ends in death without any ransom recorded. Kidnapping in Nigeria is best read as a violent crime within which a ransom market operates, rather than a ransom market that occasionally turns violent.

Source. The corpus is built from scraped Nigerian news coverage, filtered to kidnapping-relevant reports and updated continuously from monitored global news indexes. Collection keys on kidnapping, abduction and hijacking terms — not ransom — so ransom is one observed outcome among several rather than a selection criterion. Each record is a media report, not a verified incident; the same event may be reported more than once.

Extraction. State, perpetrator type, location type, outcome flags (killed, released, rescued), and ransom signals are extracted by pattern matching against the report text. Enriched fields — incident date parsed from relative phrasing, specific location, perpetrator status, and state response — are extracted where the text supports them and left blank otherwise. Items that cannot be supported from the text are recorded as absent rather than inferred.

Live layer and review. New reports are ingested automatically and flagged for human review before they are treated as confirmed. An editor verifies or corrects each machine-extracted record in the content-management system. Victim names are held only in the review system and never appear on any public surface, in keeping with the child-safety and privacy commitments of the project.

Three caveats specific to this data. Mediated voice: where a victim or perpetrator is quoted, the quotation reaches the record through journalists, officials or handlers, and should not be read as unmediated testimony. Propaganda versus data: statements issued by armed groups are strategic communication, not neutral event data, and are treated as such. Silent-majority bias: the record over-indexes on high-profile cases and under-counts rural, low-visibility incidents where no journalist was present. These limits are structural and are stated here so that users weigh the figures accordingly.

Geography. State boundaries use geoBoundaries ADM1 (CC BY 4.0). Incident points are placed at state level with deterministic spread; individual point positions are indicative, and state-aggregated figures are the reliable spatial signal. The police-proximity overlay uses an OpenStreetMap / LGA divisional proxy and is indicative rather than authoritative.