Primer: The Difference between Retrodictive and Predictive Statistics
Given that it is Pitching Week here at RO Baseball and things are about to get a little stat-heavy, we thought that it would be in everyone’s best interest to explain the difference between retrodictive and predictive statistics.
Retrodictive statistics attempt to measure the past. Predictive statistics attempt to predict the future. It’s that simple.
The classic example to demonstrate the difference between retrodictive and predictive statistics is standings versus power rankings. Assuming that teams play a balanced schedule, standings would be the best way to measure the quality of each team in a given season and, thus, would be the better retrodictive statistic. On the other hand, power rankings attempt to predict the quality of each team moving forward and, thus, would be the better predictive statistic.
For instance, the Cubs have the best record in baseball right now, so it could be reasonably concluded that they have been the best team in baseball up until this point. However, if the Cubs lost all their current players to injury, they would most likely not be the best team moving forward. The Cubs’ loss of players would not be reflected in the current standings but would be reflected in the power rankings.
Here’s where it gets complicated. Almost every baseball statistic is both retrodictive and predictive with varying degrees of validity depending on what you are trying to measure/predict.
For example, strikeouts per nine innings is a great retrodictive statistic for measuring how often a pitcher struck out opposing batters over a given period of time. However, strikeouts per nine innings is not the best retrodictive statistic for measuring how productive a pitcher was during a given time period. Other statistics, such as ERA, are better at doing that.
That being said, ERA is not the best predictive statistic for predicting future ERA. That is because ERA is affected by entities, such as the quality of the defense, that are out of the pitcher’s control. Statistics, such as FIP, over which the pitcher has more direct control would be better predictors of future ERA.
As you read the various articles that will be published during Pitching Week, consider what the author is arguing. Is he arguing that a certain statistic is good/bad at measuring past pitching performance and is, thus, a good/bad retrodictive statistic? Or is he arguing that a certain statistic is good/bad at predicting future pitching performance and is, thus, a good/bad predictive statistic?
