At the moment I write this, the Brewers are 19-7, having won 73% of their first 26 games. Brewer fan reactions tend to be some combination of at least these two emotions: (1) absolute delight at this ridiculously-good start, and (2) confusion as to what this hottest of starts tells us about the Brewers as a team.
This confusion is understandable because, to put it politely, the Brewers were not exactly pre-season darlings this year. Of the slews of writers at the baseball analysis sites, the Brewers were picked as a Wild Card by two writers at Baseball Prospectus, one writer at Fangraphs (@jasoncollette, who recently joined us for a podcast), and one writer at CNNSI. Otherwise, by what I would estimate to be at least a 10-1 ratio, the remaining analysts had the Brewers penciled in for the couch in October.
So, we have a bit of a dilemma as to how we harmonize that which the Brewers were generally expected to be (bad) and how they have played so far (extraordinarily well). Many of these dismal preseason projection were reasonable, but nonetheless they appear to have been off. At the same time, the Brewers have several months to go, and certainly almost anything can happen. But what is most likely to happen?
One thing you should NOT do is to adopt either of the more extreme positions that seem to be raging around these days:
- The Brewers will eventually lose a bunch of games and be a 74-win team, just like last year, because it is early and they will “regress”;
- The Brewers have proven their detractors wrong and are a post-season team until someone shows otherwise;
Neither perspective is likely to be correct, and the reason is that they are taking the wrong approach.
I hope you all have read Nate Silver’s The Signal and the Noise. If you haven’t, please rectify that immediately. In the book, Silver explains why society needs to move away from “all or nothing” thinking, and instead take a more scientific (Bayesian) approach to making their predictions. A scientific prediction, in Silver’s opinion, requires you to consider both (1) the prior probability you had assigned to an event occurring; and (2) the most recent data you have as the event unfolds. The key is to understand that neither your prior prediction nor the data you have coming in perfectly represent the truth of the matter. However, both play an important role. Your previous prediction needs to be updated to account for current events, but you need to ground yourself with the previous prediction to lend some context to what you are seeing and avoid changing your opinion every day.
Fortunately, for a baseball seasonal prediction, we don’t need to run elaborate Bayesian models. A simple weighted average will do, one that provides greater weight to the original pre-season prediction at first and gradually more weight to the actual game results as the season progresses. By the end of the year, the prior prediction will be almost completely subsumed by actual results. But until then, it constrains and lends context to what we are seeing on the field. The formula, which you should feel free to forget, looks like this:
Predicted Final Season Wins = (Actual winning %) * (Games played / 162) + (Predicted winning %) * (Games remaining / 162)
Let’s put this into practice. Here is a chart of different preseason predictions from various sources. I’ve simplified it so you can focus on the inputs and the results: