Change the counter. Change the rule. Change the era. One measurement refuses to budge.
This is the chapter of the series that needed the most care, so the rules come first. Everything below is a measurement, quoted from each counting project's own published figures, under that project's own rule. Where researchers disagree about what the measurement means, the disagreement is named instead of settled. There are no verdicts here about any officer, department, or policy, and no cause is asserted. What this essay will not do is pretend the measurement does not exist, because a series about who counts what cannot skip the most consistent number in the whole record.
Take the three counters in this series that publish rates by race, and notice that they share almost nothing. One counts only shootings, one counts all killings, one is a forty-year statistical reconstruction. One is a newspaper, one is a nonprofit, one is a peer-reviewed global health study. They cover three different spans of time. Here is what each one, separately, found.
| Instrument · rule · era | Black rate vs white rate | Ratio |
|---|---|---|
| The Washington Post · fatal shootings only · 2015-2024[src] | 6.1 vs 2.4 per million per year | 2.5x |
| Mapping Police Violence · all killings · 2024[src] | published as a likelihood ratio | 2.9x |
| GBD 2019, The Lancet · all police violence, reconstructed · 1980-2019[src] | 0.69 vs 0.20 per 100,000, age-standardized | 3.5x |
This series has spent eight essays showing how much the answer changes when the definition changes: about 110 people a year appear or vanish on one counting rule, and whole years go dark on another. Against that backdrop, the table above is remarkable for the opposite reason. The disparity survives every definitional choice this series has cataloged. Count only shootings, it is there. Count every killing, it is there. Reconstruct four decades from death certificates and statistical modeling, it is there, and largest. The Lancet team's full breakdown puts the age-standardized rate at 0.69 per 100,000 for Black Americans, 0.35 for Hispanic Americans, 0.20 for white Americans, and 0.15 for other groups. A number that no definition can move is not an artifact of anyone's definition.
Here is the honest boundary, and it matters as much as the table. These are population-level rates: deaths divided by how many people of each group live in America. Researchers who study this argue, at length and in good faith, about what the right denominator is. Divide by population and you measure the burden each community carries, and the ratios above are what you get. Divide instead by police encounters, or by arrests, and you are asking a different question, how encounters end once they begin, and the answer changes because encounter rates themselves differ by race, for reasons that are themselves part of the argument. The choice of denominator is the entire academic fight, and this series does not pretend to settle it.
Now put this essay next to essay six, because this is the part the rest of the series uniquely earns. The Lancet team also measured whose deaths the government's own records failed to count, and the answer was not evenly distributed: death certificates misclassified 59.4 percent of Black Americans' deaths from police violence, the highest share of any group, against 55.5 percent overall.[src]
Those are two separate failures of two separate systems, and they point the same way. Black Americans die from police violence at the highest rate of any group, and when they do, the official record is the most likely to file the death as something else. A count that is missing more than half of a community's deaths cannot even describe that community's problem, let alone inform an answer to it. That is this series' thesis wearing its heaviest clothes: legibility is not distributed equally, and the people the state sees least clearly are the people carrying the most.
What this essay has said: three independent instruments, under three different rules and eras, each measure Black Americans killed at 2.5x, 2.9x, and 3.5x the white rate at the population level, and the official undercount is worst for the same group. Those are quoted measurements, re-derivable from each project's published figures.
What this essay has not said: why. The causes are the most contested question in the entire literature, entangled with geography, poverty, gun prevalence, crime exposure, patrol patterns, and the encounter-stage decisions the denominator debate fights over, and honest researchers using different benchmarks reach different conclusions about where in the chain the disparity enters. The series' discipline holds here or it holds nowhere: measure first, attribute carefully or not at all.