Playing with quantiles, part 2
[This article was first published on Freakonometrics - Tag - R-english, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.
It is common to look at best time at the Marathon. Or perhaps the distribution of the top100, as done by John Myles White on his blog here (data can be found there), as the graph below, with the density of the time for the first 100 men (in blue) and the first 100 women (in red).
Hence, men and women are different and men run faster than women. So what…?A lot of people run those marathon. Almost 40,000 ran (and finished) the one in New York City. Instead of 100, shouldn’t we keep more runners, e.g the top 5% ?
The other point is that it might be interesting to correct (or control) with the age. If men and women running the marathon do not have the same age, it might be an explanation for that difference.
Unfortunately, to run a quantile regression properly I need the entire dataset, with 40,000 runners. And on the website of the NY Marathon (here) it is possible to extract some information, but not the entire dataset (and my standard R robots did not work here). So I managed to extract the “worst” 10% runners in 2008, and run a quantile regression to get the 95% quantile regression curve, using the idea mentioned here.
I got the 95% quantile curve for men (in blue) and women (in red), and similarly for top runner, namely the 5% quantile curve.
For those willing to play, here is the code (it is dirty since my dataset was dirty)
dataend=read.table("/home/arthur/NY2008-top.csv",header=FALSE,sep="t") nrow(dataend) I=dataend[,5] I[is.na(dataend[,5])]=dataend[is.na(dataend[,5]),6] q=which(substr(as.character(I),1,1)%in%c("M","F")) I=I[q] age=as.numeric(substr(as.character(I),2,3)) sex=substr(as.character(I),1,1)=="M" tp=as.character(dataend[,45]) i=which(tp=="");tp[i]=as.character(dataend[i,44]) i=which(tp=="");tp[i]=as.character(dataend[i,43]) i=which(tp=="");tp[i]=as.character(dataend[i,42]) tp=tp[q] temps=((as.numeric(substr(tp,1,1))+as.numeric(substr(tp,3,4))/60)*26.21875)/60 propend=length(tp)/n2008Men base=data.frame(age,temps) base=base[sex==TRUE,] reg=rq(temps~bs(age,7),data=base,tau=.05/propend) u=seq(20,80) bu=data.frame(age=u) y2008m=predict(reg,newdata=bu) propend=length(tp)/n2008Women base=data.frame(age,temps) base=base[sex==FALSE,] reg=rq(temps~bs(age,5),data=base,tau=.05/propend) u=seq(20,80) bu=data.frame(age=u) y2008w=predict(reg,newdata=bu)But looking at archives (here 1983 compared with 2008) I noticed something strange. First, if we compare number of runners (here starters),
Men |
Women |
|
1983 |
12,838 |
2,553 |
2008 |
25,669 |
13,163 |
it looks like more and more women run
the marathon nowadays. So this might have an impact on the results,
Anyway, this is now really a statistical analysis, but more a “how to say something when you cannot access a database entirely“. But I still hope someday I will…
To leave a comment for the author, please follow the link and comment on their blog: Freakonometrics - Tag - R-english.
R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.