Academic Identity
I am assistant professor of Sociology at Barnard College. My book (and my dissertation research) is a comparative study of technology and futures trading, an ethnography of open outcry and electronic traders. My current research is on how art specialists price cultural commodities, particularly how categories and commensuration work in the secondary/resale fine arts market. I teach courses in economic sociology, organizations, and gender.
Professional Identity
I occasionally consult, focusing on organizational change, the future of technology and financial markets, and environmental markets. I do strategic assessments of markets, technology and organizational design, with qualitative and quantitative components. If you are interested, please email me.
Personal Identity
I grew up outside Chicago, and went to school(s) at Wesleyan University, USC, and Northwestern University. I currently live in New York, with a partner who is a marketing manager for an educational nonprofit. I love movies, like to cook, and I can do a mean lindy swing out. I am INTP.
January 15, 2008
Data mining, airlines, precursors
Filed under: Technology — Peter @ 3:23 pm
Via the Washington Post comes an interesting article on data mining and the airline industry. Apparently, airplanes are not crashing enough for the airlines to be able to determine the sources and causes of accidents. That is, there is not enough variability in the outcomes (the last crash was August 2006) to do forensic analysis.
Instead, airlines are turning to ‘precursor’ anlayses, data mining a whole slew of events that have not led to accidents: unstabilized approaches, pitch rates at takeoff, pilot scheduling. The article suggests but does not detail the sheer number of variables and flights being analyzed, saying that Southwest has ‘mined data on more than 1 million flights’, but not really talking about what that means.
Organizationally, this is fascinating because so much more often we see organizations respond to events rather than trying to predict them. Or rather, as the vice chairperson of the NTSB put it, mining for precursors is like “‘reading tea leaves’ because it can require imagination to tie together incidents that don’t seem hazardous at first blush.” Arguably, it’s the imagination part that is so tricky in seeing what to make of precursors to mistakes and accidents. Even if you find them, often precursors only matter when they happen in conjunction (ie. in systems that are tightly coupled). So you can actually imagine a series of events that still would not result in a crash unless those events were temporally and organizationally tied together.
I would say that this is what we’re seeing now in the finance world, but it’s not. It’s worth another post, but there we’re seeing deliberate profit-seeking and many (though not nearly all or homogeneously) firms knowing that things could blow-up but not really caring.
h/t: Paul Kedrosky
2 Responses to “Data mining, airlines, precursors”
Leave a Reply
This site is hand-woven, and heavily borrows from the wonderful blueprint framwork. Rock on, grids!
Not quite
, but getting there..
January 15th, 2008 at 3:35 pm
what’s the difference, sir, between a contingent event analysis and, say, an insurance estimate for a piece of art?
January 15th, 2008 at 9:32 pm
Good question. They’re not equivalent for a few reasons (though I may be answering your question without answering what you are asking). For one, insurance estimates are about restitution, while contingent event analysis is about prevention. There is some prevention in insurance contracts to mitigate moral hazard, but the impulse in that field is to find replacement value.
In the contingent event analysis, the aim is prevention or somesuch - so, how can we understand why people commit violence (or better, don’t commit violence)? If 6 factors in conjunction lead to violence, or an airplane crash, how do we determine what these are, which is most important, and how they are temporally connected? Especially in the case where airplanes don’t crash, or violence is not committed, how can we determine before the fact what is going to continue to lead to a non-event? It’s an evaluation of an uncertain/ambiguous outcome (and in this sense, similar to insurance estimates), but it seems to be a different problem.
Is this as confusing for you to read as it is for me to write?