Worries about Inference: A Closer Look at the Family Research Council’s Report on Homosexual Assault in the Military

by Joshua Tucker on May 27, 2010 · 12 comments

in Uncategorized

The Family Research Council has released a report on Homosexual Assault in the Military that is now getting picked up in the mainstream media. The central point of the report seems to be new data analysis that shows that 8.15% percent of sexual assaults in the military in Fiscal Year 2009 were of a same sex nature (p.6), the vast majority of which involved two men (see footnote 4). The report also cites data showing that 2.8% of men and 1.4% of women in the US general population identify themselves as gay, lesbian, or bisexual. On this basis, the report claims that “This suggests that homosexuals in the military are about three times more likely to commit sexual assaults than heterosexuals are” (see p.1).

This is a bold and provocative claim, and therefore ought to be subjected to serious statistical scrutiny. The first step would be to validate their coding of the 1643 reported sexual assaults. The cite this DOD report as their source; if anyone is interested in attempting to replicate their coding I would be happy to post the results here. However, for the moment, let’s just posit that the claim that that 134 of these assaults were indeed same-sex in nature is correct, and instead turn to the inference made by the authors of this report: that homosexuals serving in the army are three times more likely than heterosexuals to commit sexual assaults. What are some factors that could potentially confound this inference?

  1. First and most significantly, the study provides no evidence of the proportion of same-sex assaults that are committed by homosexuals. This is crucial to the study, because the authors want to leverage the information in the study to argue that homosexuals should not be allowed to serve in the military. But their data measure assaults by men against men or women against women, not the number of assaults by homosexuals. Thus without any understanding of the proportion of same sex assaults that are committed by homosexuals, the inference that homosexuals are more likely to commit sexual assault is invalid.
  2. Second, we don’t know if the proportion of homosexuals in the military matches the proportion in the general population. The authors of the study assume that the proportions are similar, but if homosexuals are overrepresented in the military relative to the general population, then the inference is invalid. Moreover, it is not even clear that the general population is the right reference group; the military is overwhelmingly made up of young men. So even if we think the demographic composition of the military reflects the general population – which it may very well not – we’d still want to know something like the prevalence of homosexuality among 18-30 year old males, not among the population as a whole.
  3. Moreover, even if we assume that the proportion of homosexuals in the military mirrors the proportion in the general population, the conclusions of the study are dependent on a low estimate of homosexuals in the general population (<8.15/3, or <2.7%). Other studies have found higher estimates.

There are other more detail oriented data-related questions to worry about, such as how the results would be affected if a single individual committed multiple assaults or how the authors of the study coded cases where the accused was not found guilty of misconduct (see for example USMC Case #61). But overall, any one of the three big concerns listed above should raise serious questions about the study’s conclusion that homosexuals serving in the army are three times more likely than heterosexuals to commit sexual assaults, taken together they ought to lead us to approach this claim with a very healthy degree of skepticism.

One other point I can’t help making – the report also provides a bunch of (fairly explicit) descriptions of same sex sexual assaults (not sure why these needed to be in the report?), and in one section goes on to point out that a Persian-Farsi linguist was accused of committing a same-sex sexual assault (see p.14). The heading of that section: Do we really need more homosexual linguists?. So while we’re on the subject of inference, I assume from this argument that the Tailhook Scandal would lead the Family Research Council to conclude that we no longer need more heterosexual Marine Corp aviation officers?

[Hat tip to Neal Beck and Dan Kselman.]

{ 12 comments }

Comments on this entry are closed.

Previous post:

Next post: