[samd]

The supervisor must endorse the analyst's "reasonable belief," defined as 51 percent confidence, that the specified target is a foreign national who is overseas at the time of collection.

US citizens make up less than 50% of the world population. So given any target I can be more than 51% confident that they are not a US citizen, knowing nothing about the particular target whatsoever.

[WestCoastJustin]

The United States makes up 4.46% of the world population [1], so there might be reason to believe that 95% of communications in transit are foreign. If you look at the users of facebook, there are more who are foreign (non-US) that US citizens [2].

[1] http://en.wikipedia.org/wiki/List_of_countries_by_population

[2] http://en.wikipedia.org/wiki/Facebook_statistics

[hypersoar]

The 51% threshold sounds to me like something set by some manager(s) who didn't actually know anything about statistics or probability.

[Nimi]

Yep. If all analysts would target addresses with exactly 51% confidence and no higher, and their confidence is the exact statistical probability they're foreign, that would mean 49% of targets are US citizens.

[PavlovsCat]

Am I missing does something, or does "threshold of 51%" sound awfully like "false positive rate of 49%"? Imagine that with a spam filter. Just wow.

The only way that makes sense is that they whitelist the people they like, and simply don't give a flying fuck about anyone else, American or otherwise. Spy first, deny it later.

[jamesaguilar]

Only if you have no prior information. However, since the target has to be a specific person, and you have to have some reason to want to monitor them, you would have to have a good deal of prior information. At the very least, you know the networks on which they can be monitored, which already introduces a much more informative prior than "is-a human". The ratio of Americans to other people in your belief network would tend to be dominated by that other prior information.

[samd]

Yes, the more you know about your target the more confident you become about their nationality. Though I don't think the target needs to be a specific person, it could just be an email address couldn't it? Given just an email address you could be more than 51% certain it belonged to a foreign national and start collecting their data.

Sure, you might have to tortuously stretch the legal wording here to justify collection of any particular target, but if there's one thing this Administration has proven adept at it's tortuous stretching of the law.

[jamesaguilar]

You are right on the topic of an email address, but even for that there has to be some context that would cause you to want to collect its information. I guess if that is a single post that states violent intentions, then if you studiously avoid any further information, you could easily hit that 51% number. Then again, once you open the email, presumably you'll quickly derive the person's location and nationality, and you might then have to close it again.

I don't know if that "tortuous reading" thing is really specific to this administration. And anyway, I'm still having trouble figuring out how this whole PRISM thing was unexpected given the laws that congress passed. It seems like a rather straightforward reading of the law to me.

[bcl]

You are assuming a representative distribution of users of whichever service they select. That may not be true.

[ISL]

These numbers can't possibly work for interceptions within the US telecom network. The fraction of Americans using the US telecom network approaches 100%, while the fraction of Liberians using it is probably much smaller.

[twelvechairs]

Who supervises the supervisors?