Skip to content
Back to search results

UCDP Conflict Data Analysis

Introduced by
Uppsala Conflict Data Program (Ralph Sundberg & Erik Melander for UCDP-GED)
Year
2013
Type
Coding and analysis of organized-violence events and conflicts
Family
Conflict event-data analysis
Font size

What does it do?

UCDP conflict data analysis is the coding and quantitative study of organized violence using the datasets of the Uppsala Conflict Data Program.

UCDP distinguishes three categories of organized violence — state-based armed conflict, non-state conflict, and one-sided violence against civilians — and codes them from the level of individual fatal events up to annual conflict dyads. The Georeferenced Event Dataset (UCDP-GED), introduced by Sundberg and Melander (2013), pins each event to a place and date, enabling fine-grained spatial and temporal analysis of where and when violence occurs.

The intuition behind the method

To study armed conflict scientifically you need a consistent answer to 'what counts as violence, by whom, where, and how deadly.' UCDP provides that answer with unusual rigor: it requires a clearly identified organized actor and at least one death, it separates wars between governments and rebels from fighting between non-state groups and from massacres of civilians, and it applies a clear fatality threshold before calling something an armed conflict. Crucially, the georeferenced version records each deadly event at a specific location and date, so instead of only knowing 'Country X had a war this year' you can map the violence village by village and month by month. That granularity is what lets researchers study how conflict spreads, clusters, and responds to local conditions.

History

The Uppsala Conflict Data Program at Uppsala University began systematic conflict data collection in the 1970s and, with the Peace Research Institute Oslo (PRIO), produced the influential UCDP/PRIO Armed Conflict Dataset (Gleditsch et al. 2002). Sundberg and Melander's 2013 Journal of Peace Research article introduced the Georeferenced Event Dataset, disaggregating violence to the event level with precise coordinates. UCDP data are now among the most widely used resources in peace and conflict research and underpin many forecasting systems.

When to use it

Use UCDP data when you study organized violence — especially civil wars, non-state conflict, and one-sided violence — with carefully validated, conservatively coded fatality and conflict measures, and when you need fine spatial and temporal resolution via UCDP-GED. It is a standard source for the post-1989 period (1946 onward for state-based conflict dyads). It is less suited to interstate disputes short of war (where COW MIDs are standard), to real-time tracking at the speed of ACLED's weekly updates, or to questions requiring actor categories UCDP does not code.

Limitations and considerations

  • The fatality threshold for inclusion can omit low-intensity but politically important violence.
  • Reliance on reported sources introduces media and reporting biases, especially in remote or censored areas.
  • Conservative coding rules can undercount events relative to broader-net datasets.
  • Annual updating means it lags fast-moving events compared with near-real-time alternatives.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Rigorous, transparent definitions with conservative fatality coding and explicit uncertainty (best/low/high estimates).
  • Covers civil, non-state, and one-sided violence that interstate-focused datasets omit.
  • Georeferenced events enable fine-grained spatial and temporal analysis.
  • Long-running, annually updated, and freely distributed, supporting cumulative and comparable research.

Applications

  • Mapping the spatial diffusion and clustering of civil-war violence.
  • Modeling conflict onset, escalation, and termination at the dyad or country level.
  • Studying one-sided violence and civilian targeting as distinct from battlefield conflict.
  • Serving as ground-truth outcomes for conflict-forecasting and early-warning systems.
  • Comparing organized-violence dynamics across regions and over the post-Cold War period.

How it works

How it works

1

UCDP classifies organized violence into three mutually exclusive types: state-based conflict (a government against another government or an organized rebel group), non-state conflict (between two organized non-government actors), and one-sided violence (an actor against unarmed civilians). This typology determines which dataset and definitions apply to a given case.

Formula
\text{type} \in \{\text{state-based, non-state, one-sided}\}
2

Each event — an instance of organized violence causing at least one death at a specific place and time — is coded from news, NGO, and other sources with the actors involved, the date, latitude and longitude, and three fatality estimates (best, low, high) to capture source uncertainty. Strict sourcing and clarity rules govern what is included.

Formula
e = (\text{actors}, \text{date}, \text{lat/lon}, \text{best/low/high deaths})
3

Events are aggregated upward into conflict dyads and conflict-years. A state-based armed conflict requires at least 25 battle-related deaths in a calendar year between the same dyad; reaching 1,000 deaths in a year classifies it as a war. This thresholding turns the event stream into the conflict and dyad units used in much of the literature.

Formula
\text{conflict} = \{e : \text{same dyad}\},\ \text{threshold} \geq 25 \text{ battle deaths/year}
4

The coded data are analyzed at whatever level the question requires — event, grid cell, conflict-dyad, or country-year — modeling onset, intensity, escalation, or termination as functions of political, economic, and geographic covariates, often with spatial and time-series methods that exploit the georeferencing.

Formula
Y_{ct} = f(\text{covariates}_{ct}) \ \text{across space and time}

Common pitfalls

  • Treating the 'best' fatality estimate as exact while ignoring the low/high range that signals uncertainty.
  • Comparing UCDP counts directly with ACLED or GTD without accounting for different definitions and inclusion rules.
  • Aggregating georeferenced events to coarse units and discarding the spatial information that is UCDP-GED's main advantage.
  • Ignoring the 25-deaths-per-year threshold, which shapes which conflicts appear and when they 'start' and 'end.'

Frequently asked questions

What are UCDP's three types of organized violence?

State-based armed conflict (at least one party is a government), non-state conflict (between two organized non-state actors, neither a government), and one-sided violence (an organized actor, state or non-state, against unarmed civilians). Separating these lets researchers study civil war, communal and militia fighting, and atrocities against civilians as distinct phenomena with consistent definitions.

How does UCDP differ from ACLED?

Both are conflict event datasets, but they differ in scope and rules. UCDP requires at least one fatality per event and applies conservative sourcing, emphasizing comparability and a long time series; ACLED codes a broader range of political-violence and protest events (including non-fatal ones), updates weekly, and covers more event types. UCDP tends to be more conservative; ACLED captures more low-level activity. The choice depends on whether breadth/timeliness or strict comparability matters more.

What is the 25-death threshold for?

UCDP classifies a state-based armed conflict as active in a given year only if the dyad produces at least 25 battle-related deaths that year; 1,000 deaths in a year defines a war. The threshold provides a clear, replicable boundary for when organized violence counts as armed conflict, though it necessarily excludes lower-intensity violence that may still matter politically.

Related methods

Sources

  1. Sundberg, R., & Melander, E. (2013). Introducing the UCDP Georeferenced Event Dataset. Journal of Peace Research, 50(4), 523–532.

    Open DOI reference
    Copied.