This article compares the treatment of Colombia in large cross-country conflict datasets with the information of a unique dataset on the Colombian conflict (CERAC). The big datasets display a strong tendency to record fewer killings than does CERAC. Moreover, when the big datasets provide annual time series on the conflict, these figures look either erratic or flat compared to CERAC’s and often move in different directions. The article also examines the criteria of the Uppsala Conflict Data Program (UCDP) for dataset inclusion and finds them considerably more restrictive than CERAC’s. The primary differences are that UCDP generally excludes attacks purely on civilians and any activiTY of illegal right-wing paramilitary groups. It is argued here that these omissions impoverish our perception of many civil wars. A calculated modified series based on UCDP methodology and CERAC raw information closes 56% of the gap between the two approaches. The remainder appears to derive mainly from a number of small events in CERAC but not UCDP, reflecting the limits of English-language press coverage of Colombia, upon which UCDP data is based. The gap with other big datasets is also closed. The dynamics of the lower-bound UCDP curve clearly resemble the modified CERAC curve, so UCDP does reasonably well on its own terms. A brief Northern Ireland case study is consistent with our Colombia conclusions. The article concludes with a recommendation for conflict researchers to prioritize the construction of more micro-datasets that will facilitate detailed studies of conflict intensiTY and its dynamics.
