Bootstrapping has long been one of my favorite statistical procedures. The nonparametric version requires few assumptions, and it shares attributes with both simulation and common Bayesian models, both of which I love. At the end of the day I wonder, why settle for a point estimate and two values for the confidence intervals when you can create a distribution and visualize your CIs? When my friend Rich first introduced me to bootstrapping, he said that if statistics had been invented in the computer age, this is where most classes would start.
In this post, I describe how I used Empirical Bayesian methods to estimate the accuracy of NBA three-point shooters. This analysis closely follows the process outlined by David Robinson in his excellent book Introduction to Empirical Bayes: Examples from Baseball Statistics, and is performed using his ebbr package in R.^ The goal is to make a reasoned ranking of the top sharp shooters, despite inconsistent and imperfect records of how often players make the shots they attempt.
Last night as I was dozing off, I had a sudden inclination to try and add to the attempts that have been made to identify the anonymous New York Times op-ed writer. I’ve had some success in the past with machine learning and stylometry. And this is one of the most intriguing authorship questions in years.
By 2:00 am I was convinced the data had singled out Mike Pence. I even started to wonder what the ethical thing to do is when one has hard evidence about the source of text whose author wished to remain anonymous.
At the 2018 RStudio conference in San Diego, my colleague Jon and I gave a talk about how we use R Markdown to quickly go from nothing, to analysis, to a branded report that we can pass off to clients. This workflow took some time to set up, but like most automation tasks, has ultimately saved us more time and headache than it cost. If you want to skip to the talk,
A while ago, the popular data journalism site 538 posted a challenging probability puzzle:
On the table in front of you are two coins. They look and feel identical, but you know one of them has been doctored. The fair coin comes up heads half the time while the doctored coin comes up heads 60 percent of the time. How many flips — you must flip both coins at once, one with each hand — would you need to give yourself a 95 percent chance of correctly identifying the doctored coin?
While planning a holiday gift exchange this week, my wife casually challenged me with a sort of tricky probability puzzle:
Sara and I were talking today and realized that we were off by one on the rotation because last year we went sledding instead of buying gifts.
It should actually be: * Sara gives to Jonny * Jimmy gives to Thurop * Amy gives to Lisa * Jonny gives to Sara * Thurop gives to Jimmy * Lisa gives to Amy
We recently wanted to brand several of our plots for publication in the local press. I looked around and found a couple suggestions on how to add images to plots, but nothing that seemed modular or customizable. My colleague reccomended the relatively new Magick package, which provided all of the functionality I needed (plus a lot more). Here is a simple example along with the code to replicate it:
library(ggplot2) library(magick) library(here) # For making the script run without a wd library(magrittr) # For piping the logo # Make a simple plot and save it ggplot(mpg, aes(displ, hwy, colour = class)) + geom_point() + ggtitle("Cars") + ggsave(filename = paste0(here("/"), last_plot()$labels$title, ".
Recidivism, the rate at which those released from incarceration return or commit new crimes, is one of society’s most difficult social problems. The official estimate is that 55% of former prisoners will return within 60 months.
Recidivism is also, I discovered, one of the most challenging things to model and understand statistically. In this blog post, I describe our efforts to build this simulation, including how we settled on some fairly basic control structures (for loops) without giving up too much in terms of efficiency and readability.
Edit, 3/28/18: RStudio just announced Python interoperability through the reticulate package. Rmd Notebooks are unbeatable, in my opinion.
Original Post: I started using Jupyter Notebooks back when they were called IPython. I even remember having to set up a virtual Linux environment because they were not available on Windows. As much as I have enjoyed their functionality, I recently switched entirely to R Markdown in an RStudio environment. Here’s why.
Kaggle is a forum for interacting with other data scientists and competing to see who can write code that will best predict features of data. It’s a way to test your skills at statistics and machine learning, and to do a lot of human learning in the process (sorry, bad pun).
When I entered the contest to categorize crimes that occurred in San Francisco, my initial goal was to do better than random chance.