The confidence bounds for the mean of a sample depend on whether the population variance is known. If yes, you use standard normal critical values; otherwise, use t with n - 1 degrees of freedom, where n is the sample size. But what if the population variance is not known, though it is related to the population mean in a known way?

Mapping the daily risk of dying or being wounded in a mass shooting by US state, using data since 2013, cases per 100M person-days.

Simulation with many draws from an appropriate probability density function is sometimes a good enough alternative to whatever it is that these Bayesian people do using MCMC.

MSSQL code for running some summary statistics on StackOverflow reputation data on Meta Stack Exchange.

A class project for Developing Data Products, on Coursera

Peer Assessment assignment #2 in Reproducible Research (https://class.coursera.org/repdata-003/)

R version of Python code published at http://blog.yhathq.com/posts/predicting-customer-churn-with-sklearn.html

A utility for making both R deployment and collaboration among R users a little easier.

A painstaking dissection of the object returned by Zelig's sim() function, with a conclusion -- maybe right, maybe wrong -- about the case of the logit.

I wrote this a while back, then had to reuse the code and found some bugs to fix. Meanwhile the Zelig team got around to fixing plot.ci() themselves, and their plot is a lot better than mine. Use that for your work, and only dig around my code if you're curious about internal data structures in Zelig.

An example of Census data munging.

Are you a special interest blowing your hard-earned cash on expensive lobbyists? There is a better way.

Using the Wilcoxon rank-sum test to detect the risk of ill-representation to American voters who would have liked the Amash-Conyers amendment to pass.