The tricky bit about the NOAA affiliated survey was that the debrief was mandatory. The underlying justification for the whole study was that they were finding a higher rate of "I experienced harassment" in the debrief vs. in anonymous surveys and via criminal complaints and they wanted to understand why, and how to apply a correction to the anonymous surveys to address the manifold negative pressures on reporting.

> Presumably someone who's been the victim of sexual harassment or assault would be more inclined to participate in a time consuming survey to have their voice heard.

If you read the paper, you'd see that for certain kinds of incidents, this is true. If its especially egregious harassment or physical assault, those tend to get reported more often, even through regular channels. But if its just the bog-standard hostile-workplace stuff that we all have to attend annual seminars about, the reporting rate is very low, because the victims rarely feel like anything will be done about it. And its rare that anything's done about it, because the leadership doesn't have firm enough evidence that its a widespread enough problem to even do something about. What I'm saying is that even in the light of that NSF report, the Antarctic leadership are doing the same calculus you are and saying "well, it seems pretty endemic, but we don't really have any specifics". To which I say "well, then collect some specifics".

Also, re: data from military personnel, I don't have a clear idea how to address that, except to say that if military personnel are harassing civilian staff, or vice versa, that's a pretty serious problem irrespective of location, and the DoD maintains an office specifically to address that for all situations where military personnel regularly work alongside civilian personnel. So they, too, should be doing the same thing.

I know some people who worked on the NOAA Alaska fisheries study, and they were using pretty widely accepted methods for estimating e.g. "bycatch" - the incidental catching of undesired species during commercial fishing operations, because its impossible to have 100% coverage in the moment of all fishers everywhere. In general, compliance with the Endangered Species Act is evaluated via surveys that are processed and extrapolated using these same methods. The reason the paper even got published was because it turns out that those methods are pretty novel for e.g. human bycatch.

If you look at the figures from the paper, its pretty clear that the stated variance is because of year-over-year variation, rather than absolute uncertainty of the method (which checks out, if Ass-Grab-Greg or whoever finally got fired in 2016, you'd expect the harassment rate to drop in 2017, but you can't just ignore 2016 and prior when reporting for the entire interval).