From: brian [brian@stat.cmu.edu]
Sent: Friday, May 04, 2007 6:32 PM
To: 'Gaia Bellone'; sokumura@stat.cmu.edu
Subject: feedback on talk

Someone in the room afterwards said it was the most succinct talk.  I agree,
but *ALSO* it was very clear and easy to follow.  Very nice job.

I have only a few minor comments:

- nice illustration of mixture model

- nice plots of fitted 1 and 2 component mixture models

- it would be worthwhile to try to figure out how to display what MSE=50
means. For example: Can we distinguish by eye two densities that differ by
50 in your MSE measure?

- I still don't like the LR test; I don't believe it is asymptotically
chi-squared.  Don't bother to change the paper for stat practice, but do
look at the paper I sent you, and seek out other sources on this topic.  LR
alone is generally not trusted for choosing the number of mixture
components.

- Can compare the % of cells that prefer the 2-component model across cell
types by looking at them as binomial proportions

- April started to ask for an explanation of why the %'s are so different
between looking at the models through p-values and looking at the models
through LR tests.  This is a good and important question, for your future
work in this area.

best

-BJ