Blood from a Stone II

TPMmuckraker informs

As previewed by ABC News, the report warns of over two dozen population "clusters" in the northeastern U.S. "on a path" to terrorism.

In a report to be made public today, New York City Police Commissioner Ray Kelly concludes the 9/ll attacks were an "anomaly" and the most serious terror threat to the country comes from clusters of "unremarkable" individuals who are on a path that could lead to homegrown terror.

The report by the NYPD intelligence division, "Radicalization in the West and the Homegrown Threat," plots "the trajectory of radicalization" and tracks the path of a non-radicalized individual to an individual with the willingness to commit an act of terror, multiple sources say.

"The threat is real; this is not some bogey man we are creating here. There are individuals who are proselytizing, inciting angry young men to go down this path," said [Rand Corporation terrorism expert Brian] Jenkins, who reviewed and contributed to the NYPD report.

It's not entirely clear what a "cluster" means here. ABC reports that the NYPD identifies specific "mosques, bookstores, cafes, prisons and flop houses" as incubators of jihadism, but the 90-page report seems to attempt to craft a psychological and sociological understanding of the conditions that may set American Muslims on a radical path.

my bold.

Tehre is a basic statistical problem here. The NYPD is attempting to analyze the phenomenon of USA grown Islamic terror with essentially no data on USA grown Islamic terror. This is like the NSA "data mining" effort to determine patterns of electronic communication typical of Islamic terrorists in the USA with essentially no data on Islamic terrorists in the USA (hence the II in my title).

This is really very simple -- no statistical technique can enable you to understand a phenomenon without data on the phenomenon. A huge pile of other data which you guess might be related is as useful as your guess. Data analysis can't add anything to the original guess. The fact that computers crunch numbers according to sophisticated algorithms can't make data on terrorist appear without data on terrorism.

A few USA grown Islamic radicals have fantasized about acts of terror. It is not easy to draw reliable inference from such a small sample. Also they were all comically inept. The USA may be incubating competent Islamic terrorists but there is no information whatsoever on their characteristics or experiences.

I think that people are irrationally impressed by analysis and especially analysis of massive amount of data by a computer. If the data do not include the variable of interest -- US based or US bred Islamic terrorists, the analysis can not add to the uniformed guesses behind the algorithms.

update:

Spencer Ackerman read the report so I don't have to. He finds that "Contrary to its billing, the report doesn't identify actual Muslim population clusters in the U.S. that incline toward terrorism." and that "NYPD intelligence analysts Arvin Bhatt and Mitch Silber try to construct a model, based on prominent European Muslim and U.S. Muslim terrorists and would-be terrorists, that isolates patterns indicating radicalization." That is to say that Bhatt and Silber looked for patterns in the tiny amount of relevant data. Unsurprisingly they found nothing useful and concede

There is no useful profile to assist law enforcement or intelligence to predict who will follow this trajectory of radicalization. Rather, the individuals who take this course will begin as "unremarkable" from various walks of life.

Which shows that they are admirably honest. Given the size of their data set there is no way Bhatt and Silber could have obtained results worth a Baht.

update: Wow this got a link.