Jason Jones of The Daily Show looks at the remarkable lack of self-awareness of those who are complaining so loudly that the IRS is profiling and targeting Tea Party groups.
(This clip appeared on May 23, 2013. To get suggestions on how to view clips of The Daily Show and The Colbert Report outside the US, please see this earlier post.)
Jason Jones is hit or miss when they do simple comedy skits and interviews, but this time he nailed it. ♥ it
I watched the first 90 seconds but it’s apples and oranges. (And I like TDS and I am not so upset with the IRS behavior).
But profiling at an airport for security reasons versus profiling of IRS returns is apples and oranges.
You don’t have to like or dislike profiling to see that.
Profiling at an airport for security at least has the argument, we have to do something before the aircraft takes off. And the benefit is “all the lives on board that aircraft and whatever it crashes into” (ignoring the “cost” of profiling)
Profiling IRS returns doesn’t have that immediacy, time is of the essence, exigent circumstances requirement. The IRS can, at the speed of modern computing, apply pattern matching to all such groups, and can take years to do their analyses long after the returns were filed. In fact, they can wait years and let the groups file, one, two, three or more returns. And the benefit is not “lives saved” but “treasury dollars saved”.
So it’s an apples and oranges situation.
That…was…brilliant.
“Profiling”, as currently practiced by TSA, is not only not a good way to improve security, a reasonable argument can be made that it significantly worsens security.
The budget for security measures is not unlimited (it only feels like since 9/11). Any race- (or appearance-) based profiling is going to put up so many false positives as to be functionally useless in identifying anyone who might be dangerous, because no way can those false positives all be run down without ruinous expense to the securing organization.
It also opens the door for the obvious countermeasure: recruit people who don’t fit the profile, allowing them to raise the false negative rate right alongside the false positive rate.
Like two Afro-Britons with cleavers. Or two white-looking Chechen immigrants with American accents. Or the “Black Widows”, women suicide bombers from Caucasian Chechnya. Or people from outside the profile being extorted into doing it (cf. drug mules). None of these would be caught by profiling, while the security agents waste their time chasing down false positive after false positive.
Bruce Schneier writes really well about these kinds of security issues, and on this topic has written quite a bit.
http://www.schneier.com/blog/archives/2012/05/the_trouble_wit.html
Basically, in order to profile with reasonable security enhancement as an outcome, you need to be searching for people who have identifiable characteristics which are almost exclusively correlated with the bad act you’re trying to prevent, and you need to be able to narrow that profile to so tight that there are fewer false positives than true positives. Imagining the airport security situation, there were 820 million flights taken into or within the United States in 2012. Of those 820 million people, let’s be very paranoid and say there were 1000 of them who were actual, “I’m-gonna-blow-some-shit-up-soon” terrorists. In fact, make it ten thousand.
So we have ten thousand terrorists. Let’s say we have an awesome profile, and only 0.1% of the travellers fit it. That gives us 820,000 positives, of which we know only 10,000 are terrorists, in our wildly optimistic scenario. That’s 810,000 false positives, and 10,000 true positives.
Now, in a reasonable time before the plane takes off, how do you determine whether you’ve picked up one of the 81-times-more-common false positives, or the small number of true positives? And remember I’ve rigged the numbers here to be as generous to the screeners as possible, with a greatly-exaggerated accuracy, and a wildly improbably huge number of actual terrorists, ALL of whom fit our amazingly accurate profile, so we have zero false negatives to deal with.
The numbers just don’t work. Racial profiling is a blunt instrument, and good security takes careful, intelligent reasoning, using finely-honed tools for the particular job in hand. Given the false sense of security engendered by it, I can see why TSA wants to use it -- it’d make most of the other passengers feel safer (wrongly) if they saw this behaviour happening, and showing that they’re doing something useful is their main raisin-deeter these days, far more than actually providing security that’s worth the name. But don’t mistake it for anything other than theatre, a mime-show to entertain you while you wait for your flight.
This is exactly right. The high number of false positives tend to make such profiling ineffective.
Thanks, Mano. 🙂
Technically, it comes out of Bayesian probability, but I left out the funkier stuff. It’s the “base-rate fallacy”, and it can seem counterintuitive to people. If anyone wants to look at the math behind it, check out “Bayes’ theorem” on the Font of All The 100% Accurate Knowlij in The World.
I happen to be tutoring a postdoc on statistics right now. He’s done it before, but only in Arabic, and doing math in your non-native language is always tricky. It’s fun, though, I’m learning some Arabic, and I always did enjoy doing math when I can.
This OT ramble brought to you by Thomas Bayes, Mano Singham, and the numbers “420”.
If you’d watched more than 90 seconds, you’d know that if you want to parrot the right-wing talking point of the puppet in the video, the phrase is “two completely different things,” not “apples and oranges.” I mean, really, how can you seriously defend special status for rich white people if you don’t stick to the talking points?
“raisin-deeter” -- Raison d’être?
… doing math in your non-native language is always tricky.
No question about your long post--even getting math undergraduates to believe the counterintuitive consequences of that simple calculation that is Bayes’ theorem isn’t easy--but I seriously question the assertion that I italicized above. It just doesn’t agree with my experience or that of friends and coworkers whose native and/or second languages weren’t mine. Of course math isn’t all formulas, but sketches are not bound to a language and the reasoning is typically expressed in stilted phrases that are easily learned. “Teaching” math usually equals getting the student to recreate it, anyhow.
In fact, they can wait years and let the groups file, one, two, three or more returns. And the benefit is not “lives saved” but “treasury dollars saved”.
Well, no, because “non-profits” that exist to scam the Treasury by funneling taxable political-action dollars into nontaxable “public-service contributions” represent big $$ in lost tax revenue. Let a few years go by and you’re talking real money; you’ve also missed a political campaign or two. Moreover, when these scams are successful their success penalizes political contributors who respect the laws.
Thanks, CaitieCat! Very helpful (and interesting link).
Great post, Mano. 😀
One of my exes is from Italy. English is her second language and she got a PhD in Economics at an American university; she had only lived in the US getting her PhD for about 7 years. She didn’t tell me that learning math in her non-native language was any harder than doing it in Italian.
E’en so.
Then she’s a remarkable woman! 🙂
It’s not so much the doing of the arithmetic part, as “understanding the word problem well enough to know what method to use to answer it”. English is a bitch for that: because of the way we’ve accumulated vocabulary from so many different bases, there are often several ways to say even something very simple (viz: average, mean, expected value, all end up being x-bar in various ways). Plus for an Arabic speaker, he’s got to get used to doing his word problems left to right, and learning to recognize the difference between x and chi (in a physical sense), that sort of issue. Getting the concepts right is one thing, knowing when and how to apply them, that’s the bit that gets tricky in a second language.
I don’t have a strong opinion on profiling (I would tend to dislike it, and perhaps not be a fan even if it worked) but I’m not sure I follow your ‘false positive’ argument. Surely 1 in 81 is better than 1 in 81 thousand, right? In other words, if you can restrict your attention to a smaller group which by your own assumption nevertheless includes all terrorists, surely you are better off. Yes, you are still saddled with mostly false positives, but this problem is less severe than it would have been without profiling.
Other problems, such as the real terrorists not fitting the profile, could of course remain. And since I am not too worried about plane security, and disapprove of policies that reinforce racial tension, without a very strong case for profiling I will tend to be against it. But it seems to me that profiling makes the false positives problem better, not worse.
Am I missing something?
I think the bit you’re missing is the wildly overoptimistic numbers I used, on the principle that it’s best to argue against your opponent’s best argument, not their weakest.
Realistically, 99.9% accurate profiles aren’t even good science fiction. 90% is a ridiculous overestimate. Realistic numbers tend to be down in the 40s, according to a thing I read which I don’t have time to track down (so treat that number with generous amounts of salt -- it may be off by 20% in either direction easily).
Also, I stipulated 10000 terrorists, when in fact even the CIA only has 21000 people on the no-fly list, and that’s accumulated over 12 years; actual terrorists found in those twelve years are probably less than two hundred, if we’re very generous with our definition of terrorist and “found” (many who’ve been charged were more or less completely facilitated by FBI informants). Really dangerous ones, probably less than 50, and I’m counting 9/11.
The Schneier link i provided above is really good on the topic, but any discussion of Bayes and the base-rate fallacy will get you the idea. It’s like the fifty-mission cap in World War II: in fifty missions, a 1% dead/captured rate per mission means you end up with about a 40% chance of dying (0.99^50~=0.6). When you’re dealing with large populations, a very small rate of occurrence results in a very disproportionate-looking number of instances.
tyle:
I believe you may be missing a couple of contextual factors. In the below I am using ‘you’ in the general sense (where ‘you’ is the person in charge of implementing a profiling system, not you personally).
First, time. Much of the profiling that will occur at an airport occurs when people are boarding a flight (at least, if you want to prevent shenanigans on the flight). But how long can you delay a flight because you are working passengers through your system?
Second, resources. How much money can realistically be spent chasing down every positive? How many people can be allocated to the job, whether on the spot at the airport, or attached to off-site security & intelligence agencies? Do you have space sufficient to store everyone detained for profile-triggered screening at the airport?
Third, waste from ill will. What happens when people who get caught up as false positives sue because they missed their flight, or were mistreated by screening personnel? Even if you win or settle suits for less than the litigants initially demanded, you are spending money on legal counsel and chewing up valuable legal system time. What’s more, if the profiling system is tied to a rendition system, you can end up with human rights violations, such as happened to Canadian Maher Arar, which is surely a PR disaster (outside of those audiences who assume that victims, even if cleared by formal inquiries, really were terrorists, or suggest that receiving compensation amounts to “winning the lottery”).
Yes, certainly a profile might reduce the false positive problem. But even in CaitieCat’s hypothetical, which made simplifying assumptions enormously favouring the profiling system, the scale of the false positive problem appears prohibitive, beyond simply its rate (81 false positives per true positive).