Comments on: My Personal Intro to F1 Race Statistics http://blog.ouseful.info/2013/01/11/my-personal-intro-to-f1-race-statistics/ Trying to find useful things to do with emerging technologies in open education Sat, 25 May 2013 17:29:23 +0000 hourly 1 http://wordpress.com/ By: Tony Hirst http://blog.ouseful.info/2013/01/11/my-personal-intro-to-f1-race-statistics/#comment-30128 Fri, 11 Jan 2013 01:00:06 +0000 http://blog.ouseful.info/?p=9458#comment-30128 Hi Chris – thanks for that suggestion. “Matters of independence” was another of the bugbears I should have added to my opening line! As well as rankings, I have lap time data (for a couple of seasons at least, courtesy of Ergast API), and sector times for practice and qualifying going back several years (scraped from Formula One website; I’ve also got pit times, of a sort, from that site, and fastest laps).

One thing I have been mulling over is things like podium finish, top 10/middle 7/bottom 7 in qualifying, clean side/dirty side on grid (for end of first lap position change stats) etc.

]]> By: Chris Hanretty (@chrishanretty) http://blog.ouseful.info/2013/01/11/my-personal-intro-to-f1-race-statistics/#comment-30126 Fri, 11 Jan 2013 00:21:35 +0000 http://blog.ouseful.info/?p=9458#comment-30126 One of the problems with using final placings (i.e., ranks) as outcomes is that these are (a) definitely not normally distributed, and (b) not independent of one another: if Button finishes second, no one else can. You might instead want to look at models for survival data. Survival data is often used for time to mortality, modelled as a function of various risk factors. Here, the outcome is time to completion, higher “mortality” is a good thing, and risk factors are actually factors associated with racing quicker.

The benefit of using this kind of analysis is that survival analyses are quite good at dealing with censored observations — i.e., times where we don’t know how quick a driver would have finished a race, because they had a DNF.

Googling “survival analysis tutorial R” throws up a number of useful-looking tutorials. Predicting from survival models is a bit ropier, but should be possible out of the box.

]]>