Rob Arthur’s recent piece on 538 was an amalgamation of some of the more reliable baseball prediction outlets (PECOTA, Vegas lines, and Fangraphs projections).
Together, the predicted win totals from these sites correlate well with the actual win totals of most baseball teams (r = .60). If we do a little math, we find that previous performance explains about 36% of the reasonwhy baseball teams win as much or as little as they do. Now, if we knew 100% of the reason, we could flawlessly predict the number of wins for each baseball team; unfortunately, we’ve only got 36%. The result is that something other than previous performance accounts for around 64% of baseball team’s performance1.
|Each team has played 6 games thus far with an average opponents winning percentage at 47%|
Looking at the 4 year averages, the overall trend indicates that increasing team chemistry has played some role in improving these teams’ offenses and defenses. Before you get your scientist hat on, we know these findings are preliminary: It’s a small sample of schools — perhaps these teams would be performing this well without our survey2. There’s also another 4 to 5 games left to play for each team, so things could change by the end of the year. However, it’s a nice enough trend to get us excited about the intervention. Stay tuned for an end-of-season report to see if the results hold.
1 This assumes our predictive models built using previous performances can’t improve, right? We’re capturing about 36% of the reason (or variance) of future performance right now in 2016, what’s to say we can’t get better in the future? True, the correlation coefficients have ranged from just over r = .8 to just below r = .3. So, at most we can unreliably predict upwards of 64% of the variance on a good year, which still leaves 36% unexplained. Predictive power has varied over the last 20 years, but the average guess is that previous performance only eats up around 36% of future performance. ↩
2 Each team was randomly chosen and had to be a low-performing team in past. ↩