Early-Season Stats and Small Samples: What to Trust in April

Every April someone is hitting .400 and someone else is on pace for 60 home runs. It is tempting to believe it. The data says not to.

A simple test

We took every hitter with at least 60 plate appearances in March and April, at least 300 afterwards, and at least 300 the previous season, and asked which predicted his May-to-September wOBA better: his hot or cold April, or his entire previous season?

A month of plate appearances is mostly noise — a few bloop hits or line-drive outs swing it by 50 points. A full prior season is three or four times as predictive.

What the model does about it

This is exactly why the projections use several seasons with the current one weighted only somewhat more, and add a fixed amount of league-average performance before dividing. Early in the season a player's projection is mostly his history; by September the current season carries more weight because there is more of it. The same logic covers pitchers, where strikeout rate becomes meaningful fastest and home runs allowed slowest — see Projecting pitchers with the Marcel method.