And if you are curious where to test your newly acquired knowledge try a Kaggle competition called "Predict malfunctional components of ASUS notebooks" which is a great example of survival analysis usage.
http://www.kaggle.com/c/pakdd-cup-2014
Yes! I'm a huge fan of Cam's work. I bit the bullet and put down R to work through the Bayesian Methods book with Python. Learning the language and working through the book has been a lot of fun.
I worked as a lawyer in Chicago in the late 1990s and represented a client who was on the FBI top ten most wanted list. His name is Nate Hill. He was a Chicago drug dealer.
He fled to Africa and was running a coffee business as a front. US Marshals flew to Africa, found him, kidnapped him, and put him on a flight back to O'Hare.
He was on foreign soil in a country with no extradition treaty to the US. With the help of local bounty hunters, the Marshal's captured him, put a bag over his head, and put him on a plane back.
So, isn't that like invading a foreign country and breaking the law? Or so Americans think they can do anything they like? How'd they fill if some foreign power did the same in American soil?
Based on everything we've learned recently about the government's own capabilities, I would think that null hypothesis is that being added to the list has no effect whatsoever. (Since the publication of the list is so much weaker than the government's internal capabilities, and that it is the latter that leads to arrest anyway, not some grocery clerk reporting in.)
The easiest way to see this would be to assume there's a fairly close race on average between #10 and #11 in terms of wanted-ness, and so if we knew what the shortlist was and compared the incarceration dates of the runners-up we could figure out whether there's any effect at all to the publication proper.
I'd speculate the opposite way about the list. There have been a lot of crazy stories about grocery clerks reporting fugitives. The experiment you propose would be really cool to test a hypothesis with.
> df$date_put_on_list <- as.POSIXct(df$date_put_on_list, format = "%m/%d/%Y", tz = "EST")
It's a cool blog post but that code isn't near readable to me. Long lines in a tiny rectangle hidden behind horizontal scroll don't make it any easier. It looks like a box of arbitrary characters stuffed into as little space as possible.
Although the raw link is a little easier on the eyes:
All of these secondary factors (preparations, etc.) can themselves be regarded as "random" or at least unknowable in their extent and their effects. Much like lifetimes of individuals -- having some clear predictability on an individual basis, but in aggregate, well-modeled as random variables.
To be a bit more abstract, just because something is "random" does not mean it cannot be characterized with a distribution, expected life, etc.
The philosophical/mathematical effort at quantifying randomness has been called "The Taming of Chance".
The effectiveness of preparation is not random. Charting the effectiveness of such efforts, were it possible to do so, might yield a random distribution; this does not, however, imply that "all of these secondary factors" can or should be "regarded as random" for anything more than a high-level treatment that becomes worthless in the (individual) context where it actually matters.
I tried to have a very light touch with the data. Most of what I do is non-parametric -- so in this sense, yes as an individual this analysis is only good if you believe there's some similarity between you and everyone that's ever been on the list (so a finite sample).
If it was as simple as "preparation" and "profile," then how do any of them get caught? You'd think those protections would be relatively common knowledge, at least among criminals.
What does a pervasive subculture with a long history have to do with intelligence and planning/executing? I do not think that the existence of a "pervasive subculture with a long history" is meaningful/informative.
You have an isolated group of people: how surprising is it that a social animal develops its own culture distinct from the larger population that they are forcibly segregated from?
This social phenomena has persisted for a long time: Is the length of the history indicative of some quality unique to members of the subculture or is it one of the necessary byproducts of existence of a legal system?
What does a pervasive subculture with a long history have to do with intelligence and planning/executing?
Everything. The subculture is a mechanism by which an ethic of existence (or survival in this case) is propagated. Techniques, warning signs, all of these things. In this case, it is indeed unique to these people in that the techniques may only be useful when you are on the lam.
In fact, the uniqueness of this situation seems to be exactly what you are trying to handwave away as "it's own culture distinct," but you are ignoring the word itself: subculture exists not separate from other cultures, but as a part of them, and as a part of larger ones, such as the legal system. The legal system itself has a multitude of cultures: correctional officers have an interdependency with the judiciary, but you wouldn't say that they operate by the same terms. Judges and court reporters don't usually go to the same parties, and thus the culture of incarceration and evasion of the law also operate on their own terms and by the terms of the legal system at large. As a case in point, that is.
I think it's reasonable to say the degree of preparation and profile management varies significantly between someone who stole a car and someone who masterminds attacks on tall buildings in New York.
Preparation and profile are not "simple," nor are they black and white, as what those things actually mean is massively different from one situation to the next, criminal dealings or not. Characterizing them as something that either does or doesn't happen and does or doesn't entirely cancel the possibility of capture is not useful.
Interesting article - the title however is not only misleading but disturbing. I clicked because it implied that the FBI had dispensed with the inconvenient laws and judicial system and just started killing fugitives when it found them.
I mean, they sometimes do, but that's not what the article is about. If the author is reading, maybe change "will you survive" to "until you're found".
Thanks. Yeah, I worried it might be confusing. I use survival analysis and speak about how statisticians use jargon and I wanted that to be in the title somehow.
Coincidentally, the one person I knew who made the list lasted for around 400 days. Interestingly enough, another person I knew, who was just short of the list, lasted for 7 years before turning himself in.
Written by the same guy who brought you Bayesian Methods for Hackers - https://github.com/CamDavidsonPilon/Probabilistic-Programmin...
And if you are curious where to test your newly acquired knowledge try a Kaggle competition called "Predict malfunctional components of ASUS notebooks" which is a great example of survival analysis usage. http://www.kaggle.com/c/pakdd-cup-2014