Do spring training records mean anything?

Depending on your perspective, pitchers and catchers reporting to camp can bear varying levels of significance. For some, it defies everything we know about astronomy as the Solar System suddenly begins to revolve around Jupiter, Florida. For others–like myself–Tuesday was little more than the annual barrage of quotes along the lines of: "I feel like I'm in the Best Shape of My Life™," [Player X] said. (Both schools of thought, however, mourned the loss of John Gant's luscious hair.)

Yesterday was the first full workout for pitchers and catchers, but baseball never truly feels back for me until the games themselves begin. I don't care if I spend three hours on a March afternoon watching a Twins farmhand I've never heard of face a grizzled non-roster invite. It's baseball. Cardinals baseball for the first time in months.

There is a clockwork-esque cadence to spring training. The same team preview articles. The same debates over who deserves the final spot in the bullpen. The only thing that seems to change from one year to the next are the specific names.

In spite of all that, perhaps the most sacred vernal tradition is the extrapolation of Eckstein-sized data samples. Besides the normal randomness we always observe in 30-or-so games, spring training is especially bizarre. With the game on the line, a manager may turn to an A-ball pitcher who can't throw a strike to save his life just because he needs to get some work. A veteran guaranteed to crack the Opening Day roster might sporadically tinker with his swing or devote an entire start to sharpening his curveball even if he's getting shelled as a result.

I tabulated spring training and regular season records from every year since 2006, good for a total of 360 individual seasons. When comparing winning percentages, the standard deviation was 0.091, or approximately 14.7 wins per 162 games. In other words, pure statistical variance says only 68% of teams will even perform within 14.7 wins of their prorated spring record. For context, the standard deviation for FanGraphs' win-loss projections in 2017 was less than six wins according to Brendan Miller of Cubs Insider. While 32% of teams–roughly one in every three–can expect their spring training and regular season records to differ by more than 14.7 wins per 162 games, the projections only err that badly about 5% of the time.

The light blue trendline of this graph represents what we would predict a team's regular season winning percentage to be based solely on their spring training record. With each gridline signifying 8.1 wins, note how drastically many of the dots stray from the trendline. The measly 0.048 r^2 value means only 4.8% of a club's eventual regular season performance can be explained by how it played in spring training.

That said, there was still one more test I wanted to conduct. Meaningless games in February and March might not influence a September pennant chase, but do those who dominate the Grapefruit and Cactus Leagues burst out of the gates strong once the regular season begins? Using the same 360 case studies, I paired spring training records with regular season winning percentages through April 30.

Surprisingly, I found an even flimsier correlation between spring and first month records. The r^2 value, also referred to as the coefficient of determination, falls from 4.8% to 1.1% after shrinking the regular season sample to April. Meanwhile, the standard deviation grew from 14.7 wins per 162 games to about 18.2. Needless to say, these figures seem counterintuitive. I don't have a great theory as to why, but this likely goes to show just how little predictive power spring training has.

The bottom line is that the only real "winners" of spring training are the teams that avoid major injuries. Barring a key player going down or an external acquisition (i.e. a trade or free agent signing) occurring, a club you considered an 85-win team entering spring will probably break camp still considered an 85-win team.

I know it can be tempting, but don't fool yourself otherwise.