So with all due respect to the author, its annoying that the only ranking output the program saves is the final ranking.  What I've started doing is telling it to run with 0 iterations which outputs the  minimum violations ranking.  Then I change the output filename, and run it again with a flag which allows you to specify a seed.

This way I'll be able to compare the minimum violations rankings to the expectation maximization algorithm which typically takes 4-8 iterations to converge.  But given the computational requirements, it would seem to make sense to default to save every single ranking along the way in case you ever want to look at it later.

 

On Sun, Oct 7, 2012 at 10:33 AM, Eric Purdy <epurdy@uchicago.edu> wrote:
> Let me also raise a substantive issue.  Within groups there is more
> agreement about what is status-worthy than between groups.  While I think
> the idea that there exists a status ranking across all the students in a
> school has enough truth to it to merit the work being done on it, this
> approach is also obscuring the large differences in what earns you status in
> different contexts.  A focus on the causes and consequences of status
> differences (status as defined in this paper) inevitably ignores the
> importance of groups.  Of course, there are ways of trying to address the
> existence of groups that complement the work done here, and maybe its not
> even as hard as I'm making it out to be, but you always have to draw the
> line somewhere, and at that point you start waving your hands.  I guess I'm
> just saying we have to do our best to be honest with ourselves and our
> audience about the many things that might be important but we don't know for
> sure because we can't fit them in our model.

You can model people playing different status games at the same time.
Like, say for simplicity that there are two games in high school,
being smart and being cool. Then we could try to figure out which game
each player is playing (or how much they're playing both) based on
that. It would make the overfitting problems much worse, but it's not
super-hard to formulate such a model.

Also, you could probably get some sort of independent check of such
models by giving people surveys that try to determine what strategy
people think they are following. Like, "how cool do you think you are,
relative to your friends", "how do you decide whether you want to be
friends with a new acquaintance", etc.

--
-Eric