# The original data file is in .xls format provided by on cd by Steve Heston, (Fin, U Md.)
# at the instigation of Dan Bernhardt.  It was kindly transformed into a cvs file
# by Marek Jocek.  The contortions detailed in read.R  were used to prepare a data
# frame suitable for fitting QR ratings models.

Editing of the bball.csv file was done to rationalize some discrepancies in
the team names.  To see the details of these edits one would need to do diff
on the relevant files.

For the 2004-05 season, I've hand edited the season to remove games after
March 13th which was selection sunday  -- this file is called ncaap.d and
made another file ncaat.d which is just the 64 games in the 2005 ncaa tournament.
I've also ordered the first 33 games of ncaa.d so that they correspond to the
order of the "official" bracket as it appears in 

	http://en.wikipedia.org/wiki/2005_NCAA_Men%27s_Division_I_Basketball_Tournament

The source file ncaa.R does some plotting of fitted densities for the games
of the 2005 NCAA tournament based on fitting of the model. 

The file tournament contains a simulation experiment for the tournament as it
was seeded in the standard single elimination format.

The file NCAABB20045.xls  is Heston's "better quality" 2004-5 data and
includes "totals" for the ncaa tournament games which is used in the
penultimate section of the paper.  This data was matched by eye/hand
so craveat emptor.  A file with just these totals that is matched to
the order of the data.frame G produced by ncaa.R is called totals.d.

I've archived a version of the project as of April 1, 2007 -- see directory
archive. This version was a complete revision of the whole exercise after
the discovery that the initial drafts were improperly using the tournament
games in the estimation.  So the archived version has correct figures,
but incorrect text for the paper based on estimation using only the regular
season.  The new revision will try to update the estimation after each
round of the tournament.