[jonchang]

Can you point to exactly where you believe the analysis becomes inaccurate due to the researchers' purported bias? I found the analysis quite clear and the paper was even more illuminating.

If your issue is with the "final" 1,500 number (increased from 1,250), you'll find that it's well-supported in both the blog post:

>>> Keep in mind that the Bureau of Justice Statistics report itself excludes many jurisdictions in the United States that openly refuse to share any data with the FBI. The true number of homicides committed by police is therefore even higher. Though not a true estimate, my best guess of the number of police homicides in the United States is about 1,500 per year.

And the paper:

>>> As mentioned in Banks et al. (2015), these list intersection counts only include jurisdictions that reported any data (about 70%). As such, these numbers should be interpreted as estimates of the number of people killed by police in the reporting jurisdictions. If the reporting jurisdictions are missing from the dataset not because there were truly no killings in those areas during this period, but instead because they chose not to report homicides by police, the true number of police homicides could be 30% higher than we have suggested here.

Note that 1500 represents a 20% increase over 1250, which is lower than the 30% maximum estimate presented in the paper.

[leereeves]

> If the reporting jurisdictions are missing from the dataset not because there were truly no killings in those areas during this period, but instead because they chose not to report homicides by police

That's a rather large assumption that needs to be justified.

[jonchang]

It is justified, both in the citation and the actual text. The ARD dataset is driven by self-reporting from various local agencies. The fact that the FBI's SHR data and the ARD dataset have a mismatch (that is, there are police-related homicides in SHR that are not present in the ARD data and vice versa) is proof enough that there is underreporting in these datasets!

[leereeves]

I just realized that underreporting is already accounted for in the 1,250/y thanks to the statistical analysis described in the article.

A = the number of jurisdiction-reported homicides

B = the number of media-reported homicides

M = the number of homicides on both lists

N = AB / M

Now, if jurisdiction-reported homicides are unreported by a factor of X, we can derive a more accurate figure for A by multiplying A by X. We also multiply M by X, because adding cases to list A also adds a similar ratio (on average) to the matches between both lists. And the estimate doesn't change.

N = (XA * B) / XM = AB / M

This assumes, of course, that homicides in the jurisdictions that don't report to the FBI or BJS are still reported by the media. That may not be true but if it's not true, it must be proven false.