From: brian [brian@stat.cmu.edu]
Sent: Monday, April 30, 2007 3:52 PM
To: 'acrosset@stat.cmu.edu'; 'Steve Wasik'
Subject: your final presentation

Steve, Drew,

I would like you to email me a copy of the presentation slides you
used today.

It sounds like Alison is happy that you ruled out the hypothesis that
green cells are somehow different from nongreen cells.  That is great
news.

You've both done a lot of work this semester and that's good too.

Your talk was difficult to understand and follow, for both Alison and
the other members of the class, for a number of reasons:

1. you did not get the scientific terminology right - even to the
   point that Alison felt she had to jump in to correct it.  Making an
   error at this "superficial" level undermines the audience's
   confidence that your deeper work is error-free.

2. you stumbled over basic statistical information also.  E.g. when
   Alison asked "How do I read a density plot?"; this info should be
   at the fingertips of a first year student.  Or when April asked
   about some details of your ANOVA fits at different points (residual
   analyses, exact nature of models tested...); this should be at the
   fingertips of someone who is deeply involved in the data analysis.
   Some of this I chalk up to nervousness, which I think would be
   greatly helped with more practice before presentation (see #3
   below).

3. you both seemed rushed and unpracticed (both stumbling over
   material, and choosing convenient generic or colloquial words,
   sentences, or phrases instead of deliberately chosen, sharply
   informative ones) in your presentation.  That can happen because of
   nervousness.  However nervousness, rushing, stumbling, word choice,
   etc., can all be helped by practicing a talk *a lot* before
   presenting it "live".  It doesn't matter whether other people
   practice their talks or not; what matters is that your presentation
   would greatly improve with practice.

4. graphs (especially but not only density plots) were unlabelled or
   labelled in unhelpful ways.  E.g. we do not care what binwidth was
   used in nonpar density estimation in a presentation like this; we
   care what the variable is, what units it is measured in, etc.

5. very many slides had p-values without (a) saying what statistical
   test was involved; or (b) saying what scientific hypothesis was
   being tested.  Audience lost.

6. The idea of pairing p-values for ANOVA's with density plots of data
   in the different cells was interesting and potentially very
   useful.  However not enough information was provided on the slides,
   and you were not able to clarify well verbally when questions
   arose.  So this ended up being confusing instead of helpful.  Good
   idea, now well supported by your presentation.

7. presentation of mixture models was very confusing.  I did not
   understand the use of K-S test to argue that 5-component mixture
   was better than kernel-smoothed density estimate; if this is what
   you did it is a misuse of the test (assumptions not even
   approximately met).  in the end I could not tell from your talk
   what purpose was served by fitting mixture models.

9. green vs nongreen:

   - it wasn't at all clear to me what the point of this comparison
     was, until I asked Alison. Should be celar in your presentation.

   - the negative results in the cases that there were only 1 or 2
     green cells are probably all due to lack of power.  If you intend
     to keep these results, an estimate of power would also need to be
     included.  

   - the results in the normal group where there were more green cells
     was more informative, I thought.

One other note, related to the 'mixture modeling' portion of the talk:
for unequal sample sizes per cell, if the different sample sizes per
cell were not related to biology but rather were related to operator
schedule and other external factors, I would be tempted to do at least
one set of analyses using equal sample sizes per cell (by omitting
later readings from cells with copious readings).

I am glad that Alison is interested enough in your results to want to
follow up with you in a couple weeks after you have turned in your
final report.

I hope that you are able to use the suggestions above, not only to
think about the talk you gave, but also to improve your final paper
for the course (and for Alison).

Best,

-BJ