Monday April 28, 2008
See the scan of the napkin (thumbnail, below; click on it for a bigger view).
A bunch of us were in a bar in Nashville TN, speculating on the proportion of the American population that could correctly identify the state in which Barack Obama spent most of his childhood.
See the proposed multiple choice survey question on the right; political scientists best guesses on the left. The guesses range from 11% (Larry Bartels) to 28% (Lynn Vavreck). I’m down for 18%.
Friday April 25, 2008
I am doing the Spatial Voting module at the Washington University in St Louis edition of EITM (Empirical Implications of Formal Models); June 16-18, 2008.
I will teach for 2 weeks (35 hours of instruction) at the Essex Summer School, on Bayesian analysis for the social sciences (out of my soon-to-be-complete book manuscript), August 4-15, 2008.
Wednesday April 23, 2008
1 of 3. My TA for 350C, Thomas Brambor alerted me to electoral-vote.com. It looks quite promising, especially the polling data in CSV format…!
2 of 3. Doug Rivers put me on to The Rather Difficult Font Game. I didn’t get a perfect score. Sniff.
3 of 3. Matt Levendusky alerted me to a discussion of the use of Optima by the McCain campaign in some of the NYTimes blogs. Optima is hardly a “wow” typeface, with a little bit of interest coming from the use of its bold variant in the McCain logo (see below).
The Monterey Bay Aquarium uses Optima, as did Expo 67; what more can I say…?
Me. And she won.
Then she wins Indiana, he wins North Carolina. And we stumble through to June without any real change in the balance of pledged delegates. It is tough to see how he doesn’t become the nominee, the only question is when.
Its a great testament to the weakness of the American parties that there seems to be no mechanism where this can be brought to a head: for a while I was a fan of the theory that someone like Nancy Pelosi could use whatever authority she has as Speaker to get the House superdelegates to break one way or the other, but in the end, those House members will probably do whatever they thought was going to play best with their own primary constituencies.
I was at a conference last weekend where a Republican consultant floated the theory that she’s actually gunning for the Veep slot, or at least keeping her options open. Maybe. At the same time, Republicans were joyous at the sight of this intra-Dem fight, but nonetheless pessimistic about their chances in November (no surprise there).
Wednesday April 16, 2008
Oh. Mi. God. The full horror is on You Tube. Herr Gruber’s daringfireball alerted me to this travesty.
Not bad. Streaming the debate and a 25 person dial group from the Delaware Valley.
This is from the ABC affiliate in Philadelphia.
Tuesday April 15, 2008
I am one of a number of LaTeX users on the Mac platform noticing some font-mangling by formerly reliable friends such as LaTeX Equation Editor. The issue seems to be that Apple’s Type Services’ font caches are getting corrupted along the way, and then anything that relies on them will give odd renderings of typefaces. This includes applications such as Preview.app, Skim.app, and of course LaTeX equation editor. Adobe doesn’t use ATSServer and doesn’t seem to be affected.
There has been some web traffic on this around the place: e.g., here, and a possible solution here.
The first sign of trouble is launching LaTeX Equation Editor and encountering this:
which is actually good PDF despite the appearence. Dragging and dropping this into my blogging software produced what is supposed to look like:
Still, the issue is what would happen if you dragged this Keynote:
Deleting font caches and databases and rebooting is a solution I am working with, and then staying away from LaTeX Equation Editor… The issue is what other product to use? LaTeXIt doesn’t seem to play nice with non Computer Modern typefaces… Oh well.
Update: after using atsutil to clobber the font databases and restarting the machine, LaTeXIt seems to work like a charm. See the thumbnail below.
Monday April 14, 2008
So here is something I’ve been meaning to gripe about for a while.
I’d like to have Sweave as the (default?) TeX command option when editing Rnw files. There is a little bit of material on this on the web: a suggestion that you augment the TeX-command-list with a Rnw-mode-hook. This proposed solution doesn’t seem to work for me very well: I never see the Sweave menu items in the command menu when editing a Rnw file.
So I created my own addition to the TeX-command-list:
(“Sweave” “/usr/local/bin/Sweave.sh -ld %s”
TeX-run-command nil t :help “Run Sweave and then pdflatex”)
This seems to work, although this command option is now there for all TeX files, and not just Rnw files.
A thumbnail screenshot of my Aquamacs buffer appears below: does the status bar look right when editing a file of type Rnw? This is all with Aquamacs 1.3b, which uses ESS 5.2.7 and AUCTeX 11.84.
Thursday April 10, 2008
Via Shanto Iyengar, who simply said “strong leader?”
Sunday April 6, 2008
I am working with James Holland Jones from the Anthro Department, fitting two-component bivariate normal mixtures to birth weights and gestational ages. Mixtures are kind of fun; most of my previous experience has been with univariate normal components: it is interesting to see how these things work (or don’t) in higher dimensions.
Anyway, we’re exploring the posterior density of all the model parameters via MCMC (all the conditionals are nice and easy for the conjugate setup, so its bivar-normal/inverse-Wishart and Dirichlet/binomial sampling all the way), and so we have a bunch of samples from the joint posterior of the means and covariance matrices of the two components (inter alia).
I was thinking of a way to summarize posterior uncertainty in the covariance matrices and the means, and came up with the following plot. Over something like 4,000 samples, compute and plot the 95% confidence ellipse implied by each iteration’s draw of the variance-covariance matrix and means, for each component (since we’re fitting normal components, the means and covariance matrices are sufficient to characterize each component); library(ellipse) in R does the trick. Toss these down onto a 2d plotting surface, with the alpha channel letting the color get heavier where there are more lines superimposed on one another. A similar alpha channel hack to show uncertainty in the means of each component. Also overlay the data used in the model fitting (here, a sample of 5,000 data points from California public records).
The end result is a 40M PDF file (i.e., we could probably use less than 4,000 MCMC iterates!), converted to PNG for this blog post (thumbnail below). Graph de jour…!