Running, and Trends

Almost a year ago now, I took up running, and somehow managed to stick at it, against all historical precedent. The first thing that hooked me was being able to look at my data after the event using apps on my phone. Then once that wore off, I think I had the running bug and just really enjoyed getting out and about, exploring places.

In that time, I've been working on code in R with Genaro Sucarrat and Felix Pretis to implement indicator saturation, a technique that involves adding dummy variables for every observation in a sample, in batches to avoid perfect multicollinearity. It's an econometric method that's gaining in acceptance and use - it doesn't distort inference, and helps ensure the assumptions placed on your error terms are closer to being satisfied. A look through David F. Hendry's recent working papers and articles gives plenty of reading material on the procedure.

The applications of a procedure such as this are huge; very simply it's a very effective method of outlier detection, and also for structural breaks. Split trends can be considered also, and indeed they are in a not-yet-released version of our R code.

One of the virtues of not having been running long is that I'm still on an upward trajectory, getting faster over time. This translates into structural breaks, and perhaps split trends, if my runs are to be plotted as a time series. Really good/bad runs might come out as outliers too.

Here's the actual output if I run indicator saturation on my runs:

What does it say?

It says that back a year ago (run 0), I was running at about 7 minutes per kilometer, which translates into running 5km in 35 minutes, and 10km in 1 hour 10. After run 34 I got faster, about 6 minutes per km, with a slight further improvement to 5 minutes 48 seconds per km at run 50. This held right through to run 122 when my pace shifted up (down on the graph) a little further to 5 minutes 13 seconds per kilometer, which is where I remain now.

There's three outliers, which of course I can explain - one is an ill-fated adventure with a backpack with my laptop in (running to work - I now run without laptop and use a desktop at work), the other two are very slow trots back after very long runs where I ran out of steam some distance from home.

Overall - no trends, which is probably not a bad thing, as I might start putting pressure on myself to keep a trend going, which might not be healthy...