Downloading Retrosheet data and runs expectancy

In our book, Max and I describe the process of downloading Retrosheet play-by-play data (Appendix A.1) and computing run values of all events (Chapter 5). Here we illustrate some updated functions for downloading the data and computing the run values.

Downloading the Data
Our book assumes that you have a Windows environment, but now I use a Mac laptop and so I was motivated to adapt our downloading instructions for a Mac.

1.  If you have a Mac, you need to install the Chadwick files.  Here is an excellent description of how to install the Chadwick Software on a Mac.

2.  In the current working directory, create a “download.folder”, and two subfolders “unzipped” and “zipped” inside “download.folder”.  From the book script and data web site, download the file fields.csv and put this file in the “upzipped” folder.  (For a Windows computer, you need to have the Chadwick cwevent.exe inside the “upzipped” folder.)

3.  I updated a function parse.retrosheet2.pbp (a slight modification of the one provided in our book) that downloads the Retrosheet play-by-play and roster files for a particular season and uses the Chadwick program to extract the data and creates two data files. What’s new is that the function will work for both Mac and Windows. You can see the function here — you can read this function into R by typing …

library(devtools)
source_gist(8892981)  # reads in parse.retrosheet2.pbp 

Computing Runs Values
1. In Chapter 5, I describe how to compute the run values for all plays. I have put all of the R code into a new function compute.runs.expectancy which is found here and can be sourced into R:

source_gist(8892999)  # reads in compute.runs.expectancy function

2. Now we’re ready to download, say all of the 2013 season play-by-play data, and compute the runs values.

# reads in 2013 retrosheet files, creating two new csv files
parse.retrosheet2.pbp(2013)
# move to folder containing all2013.csv, roster2013.csv files 
setwd("download.folder/unzipped")
# computes runs values and other variables for all states
d2013 <- compute.runs.expectancy(2013)

3. The data frame d2013 contains all of the 2013 play data. I wrote a short function runs.expectancy to compute the expected runs in the remainder of the inning for all 24 bases/outs situation.

runs.expectancy <- function(data){
  RUNS <- with(data, aggregate(RUNS.ROI, list(STATE), mean))
  RUNS$Outs <- substr(RUNS$Group, 5, 5)
  RUNS <- RUNS[order(RUNS$Outs), ]  
  RUNS.out <- matrix(round(RUNS$x, 2), 8, 3)
  dimnames(RUNS.out)[[2]] <- c("0 outs", "1 out", "2 outs")
  dimnames(RUNS.out)[[1]] <- c("000", "001", "010", 
                               "011", "100", "101", "110", "111")
  RUNS.out
}
# illustrate this for 2013 play-by-play data
runs.expectancy(d2013)
    0 outs 1 out 2 outs
000   0.46  0.24   0.09
001   1.31  0.92   0.35
010   1.09  0.62   0.30
011   2.00  1.37   0.55
100   0.82  0.50   0.21
101   1.80  1.11   0.49
110   1.39  0.84   0.40
111   2.17  1.56   0.72

The 2013 expected runs are similar to those found from 2011 season data in Chapter 5 of the book. In the next post, I’ll use this play data frame to see which players last season were best in performing in the clutch.