Some commentators suggest that we have the data: CIA-directed drone strikes appear to involve fewer civilian casualties (e.g., less collateral damage) on average than DOD-directed drone strikes. Having now seen another round of such claims following Ken Dilanian’s recent LA Times story, I want to ask whether that assertion is well-grounded. On my read, the evidence is more mixed.

In a subsequent post I will address a second-order question: if the CIA-directed program performs better, what explains the discrepancy?

These are obviously empirical questions that rely on limited publicly available data. So, any analysis—by any of us without hands on the actual data–needs to be appropriately cautious. At the same time, we should also be careful about deferring too much to statements made by people with access to more of the data. There are often political and institutional interests aligned with presenting the data in one way or another.

Some of the strongest evidence that the CIA program is more accurate includes specific incidents in which JSOC-directed strikes have gone horribly wrong. Two of the most salient examples are the strike on the wedding convoy in December 2013 and the strike that killed the 16-year-old son of Anwar al-Awlaki. In addition to those kinds of data points, Mark Mazzetti’s well researched book explains that the President and John Brennan assigned the CIA the role of killing Anwar al-Awlaki due to the CIA’s success rate in Pakistan (see Jack Goldsmith’s post over at Lawfare). The White House obviously had a lot at stake, and given the data at their disposal, they decided to trust the CIA with one of the most sensitive operations. That said, I wonder whether the rationale was more about the CIA taking over the Anwar file in general, not specifically related to drone operations. After all, other evidence suggests that the first attempt to kill al-Awlaki using the CIA’s mole was carried out with Yemeni ground forces. Not a drone.

Other evidence in favor of the CIA’s program is less convincing. For example, Senator Feinstein’s assertions of the better performance of the CIA-led program needs to be considered against her seriously flawed statement that “the figures we have obtained from the executive branch which we have done our utmost to verify, confirm that the number of civilian casualties that have resulted from such strikes each year has typically been in the single digits.”

Here is some of the countervailing evidence.

1. Empirical research

A recent, publicly available empirical study by Larry Lewis suggests that the CIA drone program in Pakistan has involved a higher civilian casualty rate than the Pentagon’s drone program in Afghanistan. (In an earlier post I applauded, but also critiqued, aspects of that study.) [Update: Also Greg McNeal’s in-depth examination of targeted killing programs and personality strikes states, “CENTCOM data indicate that less than 1% of targeted killing operations resulted in harm to civilians, whereas outside observers estimate that 8% to 47% of CIA strikes in Pakistan inflicted harm to civilians.”]

2. Statements by members of Congress

In contrast to Senator Feinstein’s statement, Representative Adam Schiff, who sits on the House Permanent Select Committee on Intelligence, recently wrote:

“Some Republicans and Democrats on both the House and Senate intelligence committees argue that the J.S.O.C. lacks expertise in targeting and may cause more collateral damage. But these claims are more anecdotal than evidentiary, and the intelligence committees have yet to be presented with the facts to back them up.”

Senator Carl Levin appears to share those doubts. Earlier this year, as Chair of the Senate Armed Services Committee, Levin attempted to hold a hearing with generals and CIA officials because he was reportedly “eager to dispel the notion…that CIA drone operators were more precise and less prone to error than those in the military.” The White House did not allow CIA officials to attend.

3. Government counts of “civilian” casualties

One of the conclusions in Larry Lewis’s study is that the government’s definition and counts of civilian casualties is fundamentally flawed. (Again, as a reminder, I have expressed skepticism about aspects of that study.) We do not know, however, whether those flaws, if they exist, are systematically skewed in a way that favors the CIA-directed operations (e.g., in Pakistan) compared to DOD operations. But it does suggest that the existing government “database” of civilian casualties–including the information used by President Obama and Brennan in their decisions and invoked by Senator Feinstein and others–may be unreliable along this dimension.

4. Finally, with respect to the individual cases of JSOC-strikes: we may know more about the military’s mistakes because they are more open about them.