Regression in a nutshell

To look at run prevention, I went to the same regression approach as before.  I like this because it lets us use what Doc Nerd calls parsimonious stats.  They are stats easily counted and hopefully found in the box score.  The other nice thing about regression is that the output is pretty straightforward.  It’s the y = mx + b thing that comes up in high school.  In a regression, y is your predicted outcome, and x is your predictor.  There can be multiple predictors, leading to a formula like y = m1x1 + m2x2 + m3x3 + b.  The coefficients (m1, m2, m3, etc.) tell you how to weight each of the contributing factors.

The boring part

Looking at pitching and defense ended up opening up a lot of interesting paths for studying.  These paths are arranged in layers.  Just as the bottom tier of the standings is the least interesting, looking at the foundation is a lot like looking at the Nationals.  What I found was a lot of common sense: a team that wants to prevent runs needs to (A) not give up hits, (B) not give up walks, and (C) keep the ball in the yard (Nerdstradamus: this time we used best subsets to get the order).  This simple formula will explain over 90% of the variance in allowing runs, with a high degree of confidence.  That 90% is about as good as we are with offense, and is generally considered very high.

Here’s the raw output:

Yes, that model looks like it’s missing some things to me too, but I want to hold that thought and first look at who over- and under- achieved last year.  The one major over-achiever was the Phillies.  That’ll be good news for the Mets fans that hang out around here.  The underachievers were the Rangers and the Diamondbacks.  The Rangers gave up 967 runs last year.  That is worst in MLB, and truly horrible.  The model predicts them at 913 runs, which would have been worst in MLB and truly horrible.  The model sees the Diamondbacks as a very good pitching and defense team; it’s too bad their offense is so unconvincing.

The three true outcomes?

When I look at this model, I’m surprised by the parts that aren’t there.  Namely, strikeouts.  Strikeouts show back up in the second level of the analysis, when I look specifically at what prevents hits.  But for now, teams that allowed fewer runs didn’t necessarily strike more players out; strikeouts are not a direct predictor of a teams prowess in preventing runs.

Second, errors aren’t there at all.  Team errors do not predict allowed runs at all (nor do they predict hits allowed, but that’s for another day).  This takes me to one of my favorite soapboxes.  Earned runs are not a useful statistic.  Any run is a product of the interaction between the pitcher and the defense.  The errors behind a pitcher do not give any meaningful information in how the pitcher fared.  There may be good ways to measure defense, but errors are not among them.  The case will get stronger when I look at hits.  The bottom line is that a pitcher should be uniformly punished for any run that is given up.  That penalty need not be as harsh as currently assumed for an earned run, but the pitcher bears a degree of responsibility for every run that crosses the plate.  A more enlightened approach is slightly more forgiving to pitchers for any run allowed, but indiscriminate on how that run came to pass.  Fantasy baseballers, take note.

Pitchers have direct control over the amount of HR they surrounder and the amount of free passes they give up.  The question now is how much control do pitchers have over the amount of hits that they give up?  That’s for next time.