Closer still

To continue my self-analysis of exercise routines, and specifically to follow up on my squash post, @Bain asked the following: “I'd love to see win percentage by game # within a match. I'd guess that Steve was 75% in the first two games; I was 75% in the third and fourth games, and we were even in game five.” His hypothesis sounded right to me—I also perceived that I tended to come out strong, then fade. Turns out, we were wrong again.

Here are the totals for the full set of games. I think that past the first match we've got some pretty small sample sizes and weird incentives with added on 1 or 3 game sets[1]:

But, this isn’t exactly right, since matches don’t always stretch to five games. Someone’s going to try a lot less hard in game 5 if they’ve already lost the match. In the match-relevant games[2], here is my win percentage by game:

It looks like I had a *slight* edge in game 1, Alex had the edge in 2/3 and I completely dominated Game 4 (reflecting, in part that it's often a "must-win" for me, and in part that I tend to win matches 3-1 and Alex tends to win them 3-2). A final note is that within the games that determined matches we're 44-38 in match *games* (my advantage) even though we're 10-9 in matches. I've won more matches 3-1 while Alex has won slightly more 3-2. Here's how our match scores have broken down (again, all from my perspective)[3,4]:

Put one way, Alex converts his game wins more efficiently into match wins than I do.

1. E.g., “Now that I’ve lost, let’s bet lunch that I can take game 6”*

2. (ignoring games after the match was concluded; e.g., counting only the first four of a LWLLW match -- this is why there aren't 18 games for Game 4 and Game 5)

3. There's one match missing because I recorded only that Alex won the match, not the game totals (methodologically, I also credited Alex with one game in the game totals, since you must have taken *at least* one game more than I did from the first five).

4. Also, methodologically, I’ve got two fewer match win/losses in this analysis than I did in the last one because I excluded three round robins in which we played two games against each other (result: 2-0, 1-1,0-2) for simplicity.

*Most matches included a wager on the match. A post-game Gatorade was common, as was a beer (though those rarely got redeemed). Big bets were usually lunch. Oddly, the post-match bet was often higher stakes than the first bet; e.g., a beer following a Gatorade, or the loser had to listen to a podcast of the winner’s choice or purchase and read a book of the winner’s choice.

Read more...

As close as it gets

I’ve been keeping a log of working out since 2003. For the last ~12 months, I’ve been recording the game-level detail of squash matches as well (e.g., WLLWL for a 2-3 loss). I’ve been playing @Bain since 2008, and we’ve made some jokes about how our matches have always seemed very even, but I realized this week that I have the data to test that. I did a quick look through my data and sent off this email to most frequent 2008-09 squash partner:

“So, what I have learned (in addition to the depressing stat that I've averaged 9.5 minutes/day of working out since December 2003) is that Alex and I are 11-10 in matches and 60-60 in games. That's pretty much as close as it gets. Awesome.”

Alex posted this to Twitter with the aside, “I'd thought it was much closer :-)”

Turns out, he was right. Alex had a suspicion that he had a huge advantage in the first few months we played, and that more recently I’ve taken charge. Nope – if that were true, you’d see his game total climb much faster than mine, then see me catch up. Not the case:

In fact, over 120 games, neither of us have ever been ahead by more than five games:

Not only that, but squash is notoriously streaky. If that were true for Alex and me, you’d expect wins to tend to be followed by wins and for losses to follow losses. Or, perhaps each win takes so much out of the winner that he tends to lose the next game. Nope, once again completely even. Following a win, I was 30-29. Our games are exactly as predictable as a fair coin: dead random.

Read more...

Workout length poised for a regression to pre-graduate school levels

It appears that 2004-2006, I averaged about 50 minutes per session, while in 2007 and 2008 I averaged about 20 minutes more. The most obvious driving force here is that I attended graduate school September 2006-June 2008, which left me with much more free time than when I was working.

So, what caused the longer sessions? Was it simply that I didn't have work waiting for me at home, and that allowed me to run farther? Partially, yes. I began playing squash on weekdays, something that had been strictly a weekend pursuit for me in the past. That sort of gain is likely to be reversed. Then again, having more time off also inspired me to set larger goals. I ran a marathon in the spring of 2007, and started lifting weights in spring of 2008, both of which made me spend more time at the gym. Those gains may be maintained.

A shift in the type of exercise might also help explain the change in workout length, but the exact composition and its change is unclear. It's also unclear why 2006 shows a decrease in workout length when the last 6 months were in graduate school or on vacation.

It seems clear that when predicting the future, the rise in number of workouts could be sustainable, but that the increase in workout length may or may not be in for a regression to the historical mean, depending on whether changes in exercise mix and goals or weekday squash was a larger driver of the 2007-2008 increase.

Of course, there is a significant observer/actor bias, but we'll assume that exercise is driven by stronger forces than graph creation.

Read more...