The Meaning of a Team’s 50-Game Record
The Rays and the Angels have played more than anyone, all the way up at 47 games. There’s no major reason why they presently lead the league — it’s just noise, and everyone ends up in the same place. The Cardinals and the Twins will have to catch up eventually. They’re trailing behind, having each played just 41 games. The median right now is 44. We’re through more than a quarter of the regular season.
Small samples don’t feel so small anymore. They are small, at least relative to full-season data, but now we can more safely look for trends, for disappointments and surprises. It’s true on the player level, and it’s also true on the team level. For example, check out the surprisingly good Colorado Rockies! Or, check out the surprisingly disappointing San Francisco Giants. The records are all starting to mean something.
But, just how much do they mean? I have prepared for you a quick post. Before long, every team in baseball will reach the 50-game mark. I’ve chosen 50 because it’s nice and round and, well, that’s it. If you’ve been reading for a while, you’ll notice I’ve run a post like this before. Consider this an update, with new data. How much do those early records mean? How much more or less do they mean than the projections?
To start with, here’s one plot, showing data from between 2005 and 2016. I went with that 12-year window because it’s that window for which I have preseason team projections. The projections don’t matter for this particular plot. This shows every team’s winning percentage through 50 games, and every team’s winning percentage from game No. 51 on. How predictive are the early wins and losses?
I don’t know what you were expecting, so I don’t know what you count as a strong relationship. Demonstrated here, there is some relationship. The wins aren’t all scattered at random. But the relationship strikes me as pretty weak. It’s not anything you need to analyze in isolation, though. We can compare those results to others. Here’s a new plot, keeping all the same information for the y-axis. But instead of the x-axis showing winning percentage through 50 games, I’m instead using preseason projected winning percentage.
Emphasis on “preseason.” That’s a crucial point. Our team projections at FanGraphs update every day, to take performance and injury into account. I don’t have good access to historical midseason information, so I’m leaning on the inferior preseason data instead. Nevertheless, would you look at what this yields:
That’s a much stronger relationship. It’s still far from being a perfect relationship, but this correlation leaves the previous correlation in the dust. Even though preseason team projections don’t know about major events from the first couple months, they still do a better job of predicting the rest of the year. It stands to reason that our updating projections would be even better. Maybe not by a ton, but it wouldn’t be nothing.
What’s the relative value of each data set? Let’s set through-50 winning percentage as X1, and preseason projected winning percentage as X2. Here’s the best-fit line, for coming up with rest-of-season winning percentage:
- RoS Win% = 0.038 + 0.125X1 + 0.800X2
Actual performance through 50 games isn’t worth nothing. But if you look at those coefficients, the preseason team projection is more than six times more important. To try to put it a different way, if you want to predict rest-of-season success, you use both the numbers, but the projection number is weighted more than six times more heavily. You can see how little the actual early record adds in this plot. This shows “expected” rest-of-season winning percentage and actual rest-of-season winning percentage. The expected numbers come from the above equation.
Just with the preseason projections alone, we achieved an R2 of 0.28. Folding in the actual 50-game records provides a modest boost, up to 0.30. It’s a little bit surprising, actually, how little significance there seems to be in a team’s real record through its first 50 games. I’m not manipulating the data in any way. I’m just showing you what the spreadsheet says. While I realize I work for a company that publishes baseball projections, I don’t create projections myself, so I don’t have any known bias. I’d actually prefer the real-life data to be more important than it is.
Everyone loves teams at the extremes, right? As such, here are some more numbers. First, numbers for the 25 teams with the biggest positive 50-game differences between actual winning percentage and projected winning percentage.
- Average actual win%: .623
- Average preseason win%: .480
- Average “expected” rest-of-season win%: .499
- Average rest-of-season win%: .506
And now, numbers for the 25 teams with the biggest negative 50-game differences between actual winning percentage and projected winning percentage.
- Average actual win%: .350
- Average preseason win%: .499
- Average “expected” rest-of-season win%: .480
- Average rest-of-season win%: .477
The teams up top performed a little better the rest of the way than their preseason projections. And the teams on the bottom performed a little worse the rest of the way than their preseason projections. However, the rest-of-season performances are much, much closer to the projections than they are to the actual 50-game records. If you use the equation from earlier to get expected rest-of-season records, there’s near-perfect agreement. The projections still mean a lot more.
Again, I know this is coming from a pro-projection website. And I want to make it clear that even the good projections still miss out on a lot. Projections are incredibly far from perfect, and some teams are just legitimately better or worse than the numbers think. But, so far, the three teams with the biggest positive differences between actual record and projected record are the Rockies, Brewers, and Yankees. The three teams with the biggest negative differences are the Marlins, Giants, and Mets. Of course, injuries are an issue. The Giants don’t have their best pitcher. The Mets don’t have their best pitcher. Or their best hitter! There are some partial explanations right there. Both teams should still be better, and both teams likely will be better.
The point is a simple one. There’s substance in the projections. Even more substance than there is in actual 50-game performance. Both samples of data matter, and you shouldn’t just ignore what a team has already done. But when in doubt, consider the projections first. The greatest value of early performance is really just putting some wins in the bank. That’s what the Rockies have *really* achieved.
Jeff made Lookout Landing a thing, but he does not still write there about the Mariners. He does write here, sometimes about the Mariners, but usually not.
Hey Jeff,
At what point in the season (~70 games? 100 games?) does the actual data provided by a team’s record surpass the predictive power of projections? Of course, there’s also the concern of measuring too deep into the year, because the farther you go, the less that RoS winning percentage actually means.
I was just about to comment with the same question
I haven’t looked this up but my intuition on this is never. My only hesitation on this is the trade deadline changing team composition.
The answer is definitely ‘never’. Projections will always outperform a simple team record. Even using Pythagorean wins will always outperform a team record going into the future.
I’m not entirely sure if pythag will always outperform a teams record. Because teams will make decisions based on their record in the season.
For example, pretend a .500 team is playing .565 baseball at the break. They are likely to make trades to improve making them ROS to be more like a .565 team (See 2016 Rangers). The opposite can be true where an unlucky team could be sellers making their teams talent match their actual record instead of pythag.
There might be a point that a team’s record will be a better indication of future performance than *Pre*-season projections. But a properly time weighted updated projection will include the “actual data” of the performances from that year so it should always perform better.