Companion / Counting the Killings

The overlap

Stop comparing the totals and compare the records. Mapping Police Violence tags each of its cases with the Post's ID and Fatal Encounters' ID, so the overlap of the three civilian lists can be read directly, with no guessing.

← back to the reconciliation
The finding

The civilian trackers converged

The story people expect is chaos: rival databases, wildly different numbers. The records say the opposite. By 2015 the Post, Mapping Police Violence, and Fatal Encounters had converged. In every year they overlap, more than 98 percent of MPV's killings also appear in at least one of the others. The disagreement in the totals is not disorder. It is a thin, well-understood residue.

What only MPV catches

The residue is definitional

Strip out the cases the Post shares and you are left with the MPV-only killings, about 100 a year. Split them by cause of death and the gap resolves: roughly 53 a year are non-shooting deaths (Taser, restraint, vehicle, custody) that the Post's shooting-only rule was never going to see, and the rest are gunshot deaths the Post excludes for being off-duty or that it did not log. The gap between the databases is a definition, made of about 53 people a year.

YearMPVIn the PostMPV onlynon-shootingshooting
201510999941057134
201610659591065452
201710919811104763
2018113898415444110
201910999881115061
2020116110091525993
2021114810351134568
2022120410931114566
202312481164845232
202412701174966036

From MPV's own WaPo-ID linkage, aggregated from the dataset.[src]

How many did nobody count

Where capture-recapture breaks, and where it works

The obvious next question is how many killings escaped every list. The obvious method, capture-recapture, does not work on these lists, and it is worth seeing why. Run the standard two-source estimator on the Post and Fatal Encounters for 2019 and it returns 1,092 with a confidence interval of [1,091, 1,093].

A confidence interval that narrow is not precision. It is a broken assumption. The Post's shootings sit almost entirely inside Fatal Encounters, so the two lists are nested, not independent samples of one population, and the estimate just collapses onto the larger list. We show it here as a caution, not a finding.[src]

The valid answer comes from a source that is genuinely independent of the trackers: the government's own death records. The peer-reviewed GBD study compared the civilian databases against the US National Vital Statistics System and found the official mortality data failed to attribute 55.5 percent of police-violence deaths to the police, 17,100 of an estimated 30,800 over 1980-2018.[src]

Verify

Recompute the overlap

Verify the overlap numbers

Each row recomputes a headline fact from the committed release. If a number on this page is wrong, a row here goes red.

A check did not pass. The page and the release have drifted.

The overlap is computed from Mapping Police Violence's own WaPo-ID and Fatal-Encounters-ID linkage (reproduce it with scripts/police-shootings/derive-overlap.mjs). Aggregates only, never a re-hosted roster. Full method and the capture-recapture caveat in docs/research/police-shootings-methodology.md. Data updated 2026-07-22, next review 2026-10.

← the reconciliation the essay series →

Need something like this built?

I design and ship AI tools, full-stack apps, and data pipelines — end to end, to production. Tell me the problem in a sentence; I'll give you an honest read on fit within a day.

Work with me →