A baseball hit at 100 mph and 25 degrees launch angle should travel a predictable distance and produce a predictable outcome. Deviations from that prediction can mean two things: the ballpark's dimensions, altitude, wind patterns, and wall geometry altered the outcome, or the defending team's positioning, speed, and skill altered the outcome. Both explanations produce the same residual from the expected total bases model.
The paper disentangles park effects from defensive effects using a unified regression framework. Standardized indices for each team and each park emerge from fitting home and away performance simultaneously. The key finding: estimates differ from official MLB metrics, and the differences follow consistent patterns in home versus away performance for both teams and their opponents.
The confounding runs deep. A team that plays 81 games in a pitcher-friendly park looks like a good defensive team in standard metrics — opposing batters underperform expectations. But the batters underperform because the park suppresses offense, not because the defense is exceptional. The team's road defensive numbers, stripped of the park effect, tell a different story.
Conversely, a genuinely elite defense playing in a hitter-friendly park may look average by standard metrics. The park inflates opposing offensive numbers, masking the defense's true contribution. The regression isolates each effect by requiring the model to explain both home and away performance with a single set of park and team parameters.
What you measure in a specific location is always the joint product of the phenomenon and the location. Separating them requires seeing the same phenomenon elsewhere.