Tuesday April 26, 2011
Noon talks today at Stanford:
Andy Gelman in PoliSci: “Of Beauty, Sex, and Power: Statistical Challenges in Estimating Small Effects” (PDF)
Jeff Smisek: United CEO, Noon-1pm, Cemex Auditorium:
In 2010, United and Continental Airlines merged to form the worldâ€™s largest airline, after both posted losses the previous year. The combined company has almost 87,000 employees, with union battles and layoffs expected. 10 airport hubs serve 371 destination cities in 59 countries. 5,800 daily departures carry 144 million passengers a year. They have more than 700 commercial aircraft, $40 billion in assets, and two corporate cultures to combine.
What could possibly keep Jeff Smisek awake at night?
Wednesday April 20, 2011
So the web lit up a little today with news that iPhones are collecting time-stamped location data, and in a form that isn’t particularly hard to look at (and even with some nice apps to make animated maps of your travels etc):
The database is SQLite, and I used R (and the RSQLite package) to open it up and see what is there. In my case, I found a database with 35,297 records, with timestamps ranging from June 22, 2010 to 4 days ago (the date I last synced my phone). That said, there are only 726 unique timestamps in the database, which makes me wonder what is going on. Incidentally, the time-stamps are seconds since 2001-01-01 (I don’t know which TZ): in R, I converted with:
There is a little weirdness in the data: the iPhone thinks I was in Scotland last August (when in fact I was no further north than Colchester), and a few other instances of the geo-data being off by as much as 100 kilometres or so. There are also 68 records with lat/long recorded as 0/0, almost all of which are records from when I was in England last August.
There are also fields whose names I can’t fathom, nor does a quick look suggest that any of them might be offsets to the timestamps:
 "MCC" "MNC" "LAC"
 "CI" "Timestamp" "Latitude"
 "Longitude" "HorizontalAccuracy" "Altitude"
 "VerticalAccuracy" "Speed" "Course"
Does anyone know what the fields with non-obvious names might be?
From Jose Bernardo on the Bayesian mailing list:
The Theory That Would Not Die
How Bayes’ rule cracked the Enigma code, hunted down Russian submarines, and emerged triumphant from two centuries of controversy.
By Sharon Bertsch McGrayne
New Haven, CT: Yale University Press, 2011
This is a very compelling documented account of the early times of Bayesian statistics, followed by a detailed description of modern applications. The transition from the early seventies, where Bayesians where a tiny, less than accepted minority, to the 21st exponential explosion of applications is less documented, but the book provides very interesting reading for anyone interested in the evolution of Bayesian methods.
Tuesday April 19, 2011
Randy Stevenson is putting M-Turk to work on a project to help ANES…
This is a request for you all to help me with a little experiment that may
benefit the ANES if it works out. (feel free to forward this to anyone)
What I am asking is that you inform your graduate students about an
opportunity to familiarize themselves with lots of work that has been
published over the last 20 years in the top journals in political science,
get paid for doing so, and help out the ANES.
Here is how it works: I have posted a job on Amazon.comâ€šs mTurk website
(a platform for distributing small tasks to many workers). The task is to
classify political science articles (I have about 10,000 from the leading
journals since 1980) based on their titles and abstracts into those that
take a â€žquantitative approach to a question about American mass political
behaviorâ€° and those that do not (or for which it is impossible to say
based on the title and abstract). Specific coding rules are given in the
instructions, but basically the coder reads the title and abstract and
makes a judgment.
I have put up an initial batch of 1000 articles to be coded by three
coders each. There is a qualification test (coding 4 abstracts correctly)
that you have to pass to work on the task. Right now we are paying $0.03
per classification and it takes me about 15 seconds on average (for me) to
do the classification (and you only get paid for a classification if you
agree with a majority of the three coders). We are playing around with
the pay structure. While anyone who passes the test can do this task, it
would obviously be best done by political scientists or graduate students
in political science (feel free to register and do some yourself!).
If this simple task works, we may do some more elaborate tasks that help
us identify whether articles use ANES questions, but we are taking it one
step at a time.
1. Go to https://www.mturk.com/mturk/welcome
2. On the top right click “sign in as a worker” and follow the
3. Once your registered, go to the HITs tab and search for HITS that have
the words “political science”. You should find one posted by Randolph T.
4. Take the qualification test. Read the instructions carefully!! You
must classify all four abstracts correctly to qualify to work on the tasks
(Hint just for political scientists: none of the test questions have the
answer “can not determine”)
5.Do hits – make money- become more knowledgeable – help the ANES
Sunday April 17, 2011
I can’t resist this stuff.
Friday April 15, 2011
The Duck of Minerva: Stuff Political Scientists Like #1 — Original Data… http://t.co/z8EBxRL
Let’s see how long he can keep this up…! The riff on “original data” is a good start.
#17 “Robust Standard Errors”…?
Think I’ve got the blog pushing to both Twitter and Facebook now…
Such a busy Internet…
Janet and the kids actually came up into our building to extract me from my Stanford office the other night.
On the way out Janet notices the many fliers for seminars littering the stairwell etc.
She remarks: “I haven’t been in the building for 3 years but it seems like you’ve still got the same talks”. LOL.
Thursday April 14, 2011
Audio (mp3) containing a near miss between a UAL 777 ex-SFO and a Cessna transiting SFO Class B airspace, March 2010. 480ft lateral separation and 350ft vertical. Hmmm…
That 777 must looked real big from the Cessna window.
The 777 pilot tells the controller “we need to talk”. Ya think? And apparently they did, immediately after the incident. The 777 captain also met the SFO Air Traffic Manager at SFO tower some time later.
Tuesday April 12, 2011
And the new, new post-merger United logo has been on the company’s web site for a while now, and is starting to appear on fleet livery as well.
It looks way better than the first proposal, which used the Continental typeface for the word “United”, which looked just plain bad.
This newer effort â€” slabby, sans-serif, all caps â€” does convey the authority and purposefulness implicit in the word “United”, and is less radical a departure from previous United logos. My best guess (below) as to the typeface is Trade Gothic Extended, heavied up and with some extra kerning (and I added a hint of drop shadow).