NFL Mashup with R

nfl_mashup.R

library(lme4)
library(ggplot2)
library(XML)

# grab the NFL data & compute the score difference (Home - Away) 
nfl.raw <- read.csv("http://www.repole.com/sun4cast/stats/nfl2011stats.csv")
nfl.raw$delta <- with(nfl.raw, (ScoreOff - ScoreDef))

# fit the model 
m <- lmer(delta ~ (1 | TeamName) + (1|Opponent), data = nfl.raw)

# extract the model estimates & put into a data frame
df <- data.frame(team = rownames(ranef(m)$Opponent),
                 score = ranef(m)$TeamName - ranef(m)$Opponent)                    
colnames(df) <- c("team", "score")
rownames(df) <- 1:(dim(df)[1])

# reorder the dataframe by score & add a "rank" variable
df$team <- with(df, reorder(team, -score, mean))
df <- df[with(df, order(-score)),]
df$rank <- 1:32

# need to extract the team name w/o the city, to be able to match up
# with the ESPN format
get.short.name <- function(ln)
  tail(strsplit(as.character(ln), " ")[[1]],1)

df$short_name <- sapply(df$team, get.short.name)

# grab the data from ESPN, using modified code from: http://goo.gl/mElLS
theurl <- "http://espn.go.com/nfl/powerrankings"
tables <- readHTMLTable(theurl)
espn.ranks <- tables[[1]][3:34,3]

# teams are listed in rank order, so position in the table is their rank
df$espn_rank <- sapply(df$short_name, function(x) which(x==espn.ranks))

# plot comparing the two ranking systems
g <- ggplot(df) +
  geom_text(aes(x="Homebrew Rank", y = 32-rank, label=short_name), size=8, hjust=1) +
  geom_text(aes(x="ESPN Ranking", y = 32-espn_rank, label=short_name), size=8, hjust=0) +
  geom_segment(aes(x="Homebrew Rank", y=32-rank, xend="ESPN Ranking", yend=32-espn_rank,
                   colour=I(rank - espn_rank)), alpha=.5) + theme_bw() +
  opts(axis.ticks=theme_blank()) + xlab("") + ylab("") +
  scale_y_continuous(breaks=c(-1,32), labels=c("",""))