Putting ELO on Your Radar

While trying to accomplish one research project for a blog post (that isn’t going to work) and another research project for another blog post (which I think is going to work, but is going to be hard to organize), I decided to stop writing and just look at something pretty…still tennis-related of course — my wife has already left for work.

I ran an all-surfaces ELO for each tennis week since the start of the 2017 season for the players currently in the Top 10 in the ATP rankings, and placed them on the radar chart below.  The plot only shows a player’s ELO if they started a tournament that particular week.  You read it chronologically clockwise from the top and hopefully the rest is self-explanatory.


It’s pretty interesting to look at, and shows some things I wouldn’t have noticed.  For instance, if your Top 10 players are the best marketing draws for the ATP (which is not literally true, see e.g, John Isner), you can see where the black holes are on the schedule.

I tried a different version, with the gaps filled in, using their most recent ELOs.  However, if a player missed more than four weeks between events, I started lopping 20 points off the ELO for every week thereafter (maximum 140 point deduction).  But I want to be very clear about something:  Those deductions are intended only to improve the presentation of the chart — that is, to create some space to show the other players who did not have long layoffs.  The deductions do not make a statement about the effect of the layoff.  For that, you should read this.  In particular, and fundamentally, I did not carry the deduction over to all future events.  So, for instance, a player might have a decaying ELO for a few weeks during a layoff, but his unadjusted ELO pops back on the chart when he plays again.  That wouldn’t happen if you were trying to account for the actual effects of the layoff.  So again, the deductions here are solely to create some space in the chart, not to reflect actual ELO adjustments for layoffs.

You may also like...