Overly anal data digestion, part deux

With plenty of downtime ahead of me due to a work trip to Italy, I figured it was a good time to digest some data and look for trends in my performance. I’ve been using Powercranks since July 2008 with the heaviest use from August to February. In February and March I took a rest from them due to some nagging injuries. Now, at the beginning of May I have crossed the 200 hour mark with PCs – a significant amount of time with them which should generate some changes.

Since August of last year I’ve had some good max power values in several time durations. A few of those have been repeatable while others have been one-off performances. One-offs are great for bragging rights, but I’d rather have repeatable performances to draw on for race day. With that in mind, I’ve digested my power data going back to the 2003 season and looked at it from a max enveloping perspective as well as basing my data on my top 10 or 20 performances of the year (i.e., average my top 10 or top 20 powers for various durations). The latter is a measure of my repeatability.

The plots below (for normalized and average) show that the first two years of training with power were by far my worst. I was still within the first years of training in general so the data for 2003 and 2004 are certainly expected. The key thing is that the data from 2005 to 2009 is clustered together.

Because 2003 and 2004 are outliers in the sense that I was still training to get to my peak performance, I will ignore those years from the rest of this analysis. I’m not ignoring the data from the past, but placing importance on the more immediate past (the last 5 years).

Going back to 2003 I’ve got 1120 days of power data stored. That’s 1120 data points from which to extract power data for 5 minutes, 20 minutes, 60 minutes, or whatever duration I want. Most of the data is for years 2004 to 2009, so each year has nearly 200 data points and a good sample size from which to draw conclusions.

If I envelope my top 20 performances in a year I get the data below for normalized and average powers from 3 to 180 minutes. The solid red line represents the current season since starting with Powercranks, and the dashed red line is an average of the 2004-2009 data with error bands (one deviation plus/minus). Within a given year, the deviations are on the order of 3-6 W, indicating high repeatability in my performances. For the cumulative period of 2005-2009, the deviation is on the order of 2-4 W, within the tolerance of the Powertap itself.
Notice how tightly clustered all the data is until you get beyond the 120 minute range. For durations of 2 hours and more, I simply didn’t ride hard for that long in the past. As training evolved and my race team grew, the training rides got longer and harder. The higher power experienced in 2009 for more than 2 hours is primarily due to my weekly training ride being longer.

Since I’m primarily a time trialist, what interests me more is my power for an hour or less. Zooming in on the data above for the 10-60 minute durations we have the plots below. Here 2008 sticks out like a sore thumb. It’s clearly my worst year in the last five years, but even then it’s only off by a mere 5 watts from the average and about 7-8 watts from this year and 2007. Further, 2009 looks to be a carbon copy of 2007.

If I were to instead consider the top 10 performances of the year instead of the top 20, we would expect the curves to rise a bit. And they do. The next 3 plots are for years 2007 to 2009, and I’ve plotted my yearly best along with curves representing my top 10 and top 20 values. There is a difference between selecting the top 10 or 20 values, but honestly the difference is quite small. Indeed, the major difference is between these curves and the max curve. Ideally I’d want to have all three of these lines merge. What the curves do tell me, however, is to be careful with so-called “personal bests” which could be nothing more than aberrations. Was I feeling great that day? Was it from a fresh chain or newly cleaned and lubed chain? Temperature issues with the Powertap?

Another way to look at the data is by date. The plots below are my 20 and 60 minute normalized powers for the last several years. The plot is of all dates, but only rides in the specified wattage range can be seen (in other words, rides with lower normalized power are off the graph). The 2007 training year has better consistency in the 290 W range for 20 minute power, whereas 2009 shows more consistency in the 280-290 W range. 2008 just looks like a lousy year. For 60 minute power, 2009 shows two rides over 280 W for an hour, but 2007 appears more consistent, albeit by a only few watts. Which would a rider rather have? Those two peaks that are 5-7 W higher than the cluster in 2007, or that cluster of consistency?

Finally, what if I were to take my top 10 performances of the year for various durations, average them, and using the critical power model determine my “top 10 based critical power”? I did just that, looking at data at 3, 5, 8, 10, and 15 minutes in duration. Doing this since 2003 I get a performance increase after the first 2 years and then things all blend together. My expected hour power would be 262 W for 2003, 260 W for 2004, 271 W for 2005 and 2006, 274 W for 2007, 272 W for 2008, and 273 W for 2009. When I look at the data, 2007 sticks out the most; I had my highest anaerobic work capacity contribution in that year (though honestly it’s still pretty weak). In that year I spent several weeks doing heavy VO2 work. I had access to the roads west of Paris with several shorter climbs, including one that was ideally suited for 5 minute efforts. I also utilized track racing extensively in preparation for the state time trial championships. Looking back, it appears this VO2 work was the recipe for success. Fast forward to 2009 and I’ve been doing quite a bit of VO2 work, though my 3-5 minute power is 5-10% lower than 2007.

Looking at the data, there’s several things which could be stated:

1) One-off performances are great, but unless you can back them up they are useless if done out-of-competition.
2) Consistency is more important, and by looking at the top 10 to 20 workout and race performances you get a better measure of where your fitness currently sits.
3) I’ve been remarkably consistent the last 5 years. I busted my butt in the fall with Powercranks (tempo, threshold, and VO2 workouts with them) but still I seem to be right smack dab in my physiological limits. My workouts, as measured by TSS, are up about 10% over past years with zero gain. In other words, lots of extra work for nothing (or at most one or two more watts).
4) 2008 was my worst year in the recent past, but even as bad as it was, it really wasn’t that bad, just a few percent lower than past years.