Marathons seem to be popular these days, therefore I decided to dive into some data and find a way to apply analytics. You might find this interesting if you're a fan of marathons - or if you like seeing the details behind cool graphing techniques!

Before we get started, here's a picture to get you into the mood for this topic. My friend Sunil is an avid marathon runner, and in 2016 he decided to run 1 marathon each month for a year. This Tucson marathon was his December race, where he completed his 1-year challenge. Way to go Sunil!

The Original Graph

Now, let's get going with the analysis. To get off to a fast start, I decided to check what graphs were already out there. I found a really interesting article about some of the analytics used to look for possible 'cheaters' in the Boston Marathon data. They had a graph that identified runners who ran the marathon 20+ minutes slower than their qualifying time. I decided to use their graph as the starting point for mine (and hopefully add a few enhancements along the way). Here's a screen-capture of the graph from their article:

Simple Plot Markers

I found some Boston Marathon results data on Github, downloaded it, and imported it into SAS. I created my graph by starting simple, and adding more features one layer at a time. In my first iteration, I simply plotted each runner's time as a blue marker. One small enhancement over the original graph is that I made my blue markers transparent, so you can see where the markers are 'stacking up' (multiple markers in the same location).

proc sgplot data=plot_data noautolegend noborder;
format official time_avg mmss.;
scatter y=official x=bib_num / transparency=.85
   markerattrs=(symbol=circle color=cx0000ff size=5pt);
yaxis display=(noline noticks) labelpos=top label='Race Time'
   values=('00:00:00't to '00:09:00't by '00:01:00't) grid gridattrs=(color=graydd)
   offsetmin=0 offsetmax=0;
xaxis display=(noline noticks) label='Bib Number (lower bib number = faster qualifying time)'
   values=(0 to 35000 by 7000) grid gridattrs=(color=graydd)
   offsetmin=0 offsetmax=0;
run;

Adding Average Line

In the original graph, their pink dots represented runners who had a marathon time 20 minutes slower than their qualifying time. I didn't have the qualifying time data, therefore I needed to come up with an alternative. I decided to calculate the average time for each group of 200 sequential bib numbers, and then identify the runners who ran the marathon 20% slower than the average time for their group. Here's the code I used to calculate the groups, and the averages:

data plot_data; set race_lib.boston_marathon (where=(bib_num^=.));
obs_200=round(_n_/200);
run;

proc sql noprint;
create table plot_data as
select unique *, avg(official) as time_avg
from plot_data
group by year, obs_200
order by year descending, bib_num;
quit; run;

I then overlaid a red series line on the graph, to represent this average. Notice that I used the y2axis option, to add an axis on the right-hand side of the graph - this makes it easier to estimate the times for the slower runners, and is an improvement over the original graph.

series y=time_avg x=bib_num / y2axis lineattrs=(color=red);

y2axis display=(noline noticks) labelpos=top label='Race Time'
   values=('00:00:00't to '00:09:00't by '00:01:00't) grid gridattrs=(color=graydd)
   offsetmin=0 offsetmax=0;

Identifying Outliers

I then determine which runners ran the race more than 20% slower (or faster) than the average time for their group of 200 peers (with similar bib numbers).

data plot_data; set plot_data;
if (abs(official/time_avg-1))>.2 then outlier=official;
run;

And I show the outliers on the graph using the 'x' symbol as the plot marker:

scatter y=outlier x=bib_num / markerattrs=(symbol=x color=cxff0000 size=8pt);

And here's one final enhancement in my graph, that wasn't in the original graph in the article. I added mouse-over text to the outlier (x) markers in my plot, to show the name/time/etc of the runner. Here's an example of the mouse-over text (of course, please don't assume this runner 'cheated' - I just picked this one at random to show some mouse-over text!)

scatter y=outlier x=bib_num / tip=(name country bib_num official)
   markerattrs=(symbol=x color=cxff0000 size=8pt);

 

Here's the full code I used to create the graph, if you'd like to see all the details. And if you're one of my long-term blog readers, you might think you recognize these graphs ... and you might be right! These graphs are the ODS Graphics (sgplot) version of graphs that I had originally created using SAS/Graph (gplot) in one of my blog posts a few years ago!

Just for Fun

If you made it this far, I want to reward you with a 'fun' photo. It was tough deciding which of my friends' photos to include in the blog, and here's an "also ran" photo. This is my friend Ben - he wasn't running a marathon per-say, but instead was running in a famous charity fundraiser race - the Krispy Kreme Challenge. In this race, you run from the NCSU Belltower to the Krispy Kreme store, eat a dozen donuts, and run back. This race requires some true dedication and practice!