Context Matters
The Applicability of Historical Data
Friday night in the bottom of the ninth of a tied game, the Padres had a runner on second base with one out and Luis Arraez coming up to bat. If the runner on second scored, the Padres would win the game. Historically this situation has led to 0.69 expected runs and the home team winning 72.65% of the time.
The three batter minimum rule for relievers meant that Dodgers’ pitcher Michael Grove could not be substituted until he’d faced three batters in the inning, and he’d only faced two thus far. So this left Dodgers’ skipper Dave Roberts with two potential levers to pull:
Let Grove pitch to Arraez
Intentionally walk Arraez
If the Dodgers walked Luis Arraez the base/out state would shift to runners on first and second which increases the run expectancy for the inning to 0.93 runs and the home team winning 75.84% of the time. A net negative shift of 3.19% against the Dodgers.
So when trying to determine the correct strategy, Roberts had to pick between two states:
Pitching to the batter historically meant fewer expected runs and a lower chance of the home team winning, or
Walking the batter which historically led to higher expected runs scored and a higher chance of the home team winning.
So, easy choice, right? The best strategy is to pitch to Arraez rather than walk him.
You have already probably realized the flaw here: these inferences are coming from long term averages with all of the game context stripped away. The context was: a single would likely win the game. And the hitter at the plate was the greatest singles hitter in the league. The Dodgers are an extremely data driven team and undoubtedly knew the above percentages. They elected to pitch to Arraez.
It’s interesting to wonder what their ultimate calculus was. There are any number of heuristics Roberts might have used to make his decision. The historical data suggested pitching to Arraez led to a 3.19% increased win probability for the Dodgers. But then there’s the player specific data. Luis Arraez is a career .366 hitter with runners in scoring position. The hitter after him was Fernando Tatis Jr. who is a career .324 hitter with runners in scoring position. Arraez was 4.2% more likely to get a hit than Tatis, and a hit likely wins the game. The Padres had pinch run Tyler Wade on second, one of their faster players, increasing the chances that a single would score a run to win the game. This sort of layering of context can go on and on. But the moment illustrates why late game decision making shouldn’t solely be made upon expected outcomes from large samples of historical data. There are countless other considerations to the decision. Roberts said after the game:
I just felt that, if we could pitch Arraez carefully, we get a chance to get to Tatis...You kind of pick your poison right there. Just didn't work out.
On the one hand it’s fair to say he had to pick his poison, choose between two unfavorable states and two good hitters. But the hitters were good in very different ways, and that moment aligned itself almost perfectly with Arraez’ skill set. These decisions individually only swing win probability by a percent or two. But over the course of the season that can mean an extra win or two. And 43 games into the season the Padres are 22-21, and exactly one game ahead for the final Wild Card spot. Getting these decisions right over the course of the season can be the difference between surviving the regular season and an early trip to Cancun.
On Friday the Dodgers chose to pitch to the preeminent singles hitter in the game in a moment when a single probably meant death. Irrespective of what went into the Dodgers’ decision, here is the moment replete with a masterful play call and elite situational awareness from Don Orsillo and Mark Grant who recognized that once the play by play was complete, the moment spoke for itself.
Source: @Padres