Why We (And The Projections) Hate Your Team

Regular Bat Flips & Nerds listeners will have recently heard Ben, John, Tom and myself make a series of win predictions in our long-running annual preseason game. We even let the Triple-A squad of Rob, Russell and Gav have a go in a rival podcast. If you didn’t hear those episodes, you should check them out, especially if you like yelling at your device of choice about the players on your team we forgot about, or furiously tweeting at us to explain why your favourite player is going to be better than we think.

Our own questionable predictions notwithstanding, it is the season for people getting mad about projected win totals. Whether it’s PECOTA’s evil scheme to underrate Atlanta or FanGraphs’ brutal assessment of the Orioles and their postseason chances, there’s one underlying theme to these responses: they are definitely wrong about my team.

The reality is much more complex. Let’s get one thing out of the way first: nobody, human or machine, gets everything right all the time. Even if you were incredibly accurate about 29 of 30 teams – which again, never happens – fans of the one team that you missed on by ten wins would be racing to your mentions to point out that you were wrong. So why does it feel like we, or PECOTA, or FanGraphs, or any other projection system or prognosticator, hate your team? What does it tell us about the process of making predictions, and the human condition in general?

It’s the season for optimism

Spring training might be full of meaningless games but it’s a season ripe for optimism. All the disasters of the previous season can be written off. The players who struggled through injuries are in the best shape of their life. The breakout stars are going to keep getting better. That free agency addition is definitely going to address the big issue your team had last season. Your divisional rivals got lucky last year and it won’t happen again.

Hopefully, reading that sounded ridiculous. If not, you’re probably a super-optimist. That’s a great way to approach a season. After all, where’s the fun in backing a team that you’ve given up on before the season began? One of the reasons we hate tanking so much is because it cuts against the very core of what teams ought to be doing here: trying to win baseball games. There’s little enjoyment to be found by souring the weeks of anticipation with despair. You just need to be aware that’s not the game projection systems are playing.

The projections are going to treat every player the same way, applying the same statistical rigour regardless of team. Some teams might have more players the projections “like” because of their profile – perhaps they have good plate discipline, or an exceptional strikeout rate – but they do not suffer from our tendency to overrate our own team’s prospects while diminishing or outright ignoring the flaws, or the potential of other squads.

This is not to say that every fan approaches their team this way. Some are natural pessimists, if not from the moment pitchers and catchers report then certainly from the time the team’s first significant injury arrives. On the whole, however, people are optimistic in the spring. I have been conducting various prediction and projection reviews at Banished to the Pen for half a decade now, and whenever I include a set of projections from a group of people they are, without fail, too optimistic.

What do I mean by “too optimistic”? These predictions collectively end up with more wins than are actually possible: 2430, in a regular full-season slate of games. This tends to come out at around two wins per team too high on average, which often means that on the teams that are middling-to-good, the difference is greater, because people usually are pretty pessimistic about the very bad teams. Projection systems, and our prediction game, are designed to eliminate this by forcing the win total to that 2430 mark. That brings us to the next point.

Not everyone can be a winner

There’s a fairly narrow space to occupy to be considered successful in baseball. Making the playoffs is usually the minimum bar to clear. Many fanbases would be unsatisfied if that didn’t come with a division title, unless they had a deep postseason run. A few would consider anything less than being the best team in their league to be a failure. Yankees fans might not think they’ve been successful since 2009.

Right now, there are a lot more fanbases who think their team will, or at least could be successful, compared to the number that will actually achieve their goal at the end of the season. Our prediction game forces us to reckon with this immutable rule: there will not be more than 2430 total wins (barring a game 163 scenario). You cannot simply hand out 95 wins to every team that you think has a chance to make the playoffs. This, in effect, is what is happening when you get a bunch of predictions from a series of disparate, optimistic fanbases.

If making the playoffs is “success”, only ten teams are successful. At a conservative estimate, significant proportions of at least 20 fanbases probably think they have a shot at the playoffs right now. It’s an undisputable fact that half of those will be disappointed in six months.

Take the NL Central, where plenty of fans of the Brewers, Cardinals, Cubs and Reds all think they have a shot at the division. If we’re defining making the postseason as success, at minimum one of those teams is going to fail. Given that the Dodgers and Padres are likely to take two playoff spots, that’s two failures, and if the other Wild Card comes out of the NL East, three of those teams are going to fail.

It’s easy to sit down and make rosy predictions about your team. It’s much harder when you have to hew to the 2430 rule and figure out where all the wins are coming from. We ended up with 25 wins too many during our first pass this year, and had to trim the totals of quite a few teams. I quite like the Yankees, Rays and Blue Jays, for instance, and can easily see scenarios in which they each win 90-plus games, and if the Red Sox get lucky with some pitching health and performance, they could surprise. For them all to win 90-plus games, however, you have to do some very complicated mathematical gymnastics to make it add up. At the end of the day, you have to make a call somewhere and decide what you think is most likely (for me, it’s the Yankees being the best team in that division).

Projected standings are often an unpleasant insight into a potential future that we don’t want to contemplate right now. Think about the final standings in previous seasons and how often you were happy with those. It’s the same thing. The only difference is that we were confronted with irrefutable reality, whereas we can look ahead unconstrained by such pesky factors as real life wins and losses. For now.

We split hairs way too much and fail to recognise uncertainty

There is almost no difference between a team in first and second place if only one projected win separates those teams. Two, or three, or even four – it’s simply not that significant. These teams are functionally the same as far as projections are concerned. Projection systems simulate seasons thousands of times and the number you might see on a website or a tweet is the average of those simulations. There are simulations where the Dodgers win 120 games. There are simulations where the Pirates have a .500 record!

At Baseball Prospectus, we publish graphs showing the full distributions of all these simulations along with the standings. Take the NL East, where PECOTA has made lots of people very angry because Atlanta is apparently a fourth-place team:

We can argue about whether PECOTA is too low on Atlanta elsewhere but what this says is not that they are definitely a fourth-placed team – which is what you might assume if you just looked at the average simulation total – but that Washington, Philadelphia and Atlanta are essentially about the same talent level as far as PECOTA is concerned. There are lots of these simulations in which Atlanta finishes in first; there will be some in which they finish behind even Miami. Baseball is often a random sport, with far greater uncertainty than most other major sports, and 162 games still isn’t enough to eliminate that uncertainty.

In other words, it seems very significant when you see a certain team in first or one team ranked behind another when you think they should be flipped. It’s not, especially if it’s by a win or two. The projections aren’t making some grand sweeping statement. They’re just running simulations with the available talent and these are the results. Teams who “should” win 82 games sometimes win 90. The Marlins made the playoffs last year while allowing 41 more runs than they scored. That kind of thing does level out over 162 games, but not completely.

The projections also don’t care who won the division last year. That doesn’t give a team a magic boost. All they are doing is using the projected quality of the players on the roster to determine how many runs the team will score, how many they will allow, and figure out how many games they’ll win over the course of the schedule. Pennants have nothing to do with it.

There are things the projections don’t know or can’t (and shouldn’t) account for

It’s been a rough few years for Kris Bryant. Last season it felt like he was on a quest to collect an injury to every body part. Actually, he just wanted to play some good baseball, and his health wouldn’t let him. Bryant was an MVP-level player (and winner) in his first three seasons but the last three have been a mixed bag, featuring uneven performance and injury issues.

Projection systems don’t really know that Bryant has been hurt. They see a 29-year-old whose peak was five years ago and a below-average hitter last year when he was on the field. Steamer thinks Bryant is just over a three-win player. That’s pretty good, but it’s not quite All-Star level, and it’s an awfully long way off MVP. ZiPS and PECOTA peg Bryant as more like a 2.5-win player. Again, that’s definitely a guy you want in your lineup, but if he’s supposed to be one of your three best players, you’re in trouble.

Perhaps you think Bryant is healthy and ready to get back to MVP level, or at least close to it. If that’s the case, the projections are probably at least three or four wins too low on the Cubs, a distinction that would send them flying up tightly-packed NL Central projections. You might be right. The projections, using Bryant’s statistical history, have no particular reason to believe MVP Bryant is coming back. They might also be right. More often than not, they are.

There are also players who are outliers, whose skills can’t be fully captured by a projection. Sticking with the Cubs, Kyle Hendricks is the perfect example. He doesn’t have elite strikeout rates, which projections love because strikeouts are reliable from year-to-year and they don’t leave players at the mercy of batted-ball fortune. It’s well-established at this point that Hendricks has impeccable command, arguably the best in baseball, that allows him to outperform his peripherals from a run prevention standpoint. He’s exceptional at deception, pitching around the edge of the zone and sequencing, allowing him to get superior results on balls in play. They’re also all among the aspects that are still incredibly difficult to quantify effectively.

You can’t overfit your projection system to a pitcher like Hendricks. If we went round giving elite ERA projections to pitchers with a 20 percent strikeout rate and .280 BABIPs the year prior, the projections would be dreadful, because most pitchers don’t have Hendricks’ skillset. Most of those pitchers get worse, in many cases a lot worse. There will be exceptions, like Hendricks, but if you built a system around them then it would perform very badly.

The projections really don’t care and they aren’t out to get your team

There’s a tendency to treat projection systems like a precursor of Skynet. They are at once an impenetrable black box, the logical extreme of playing the game on a spreadsheet, and the ultimate hot take generator, the perfect manipulator of your feelings since they have none of their own. Or maybe there’s some sadistic mastermind behind them, poking and tweaking them until they arrive at the configuration that makes you – yes, YOU – as angry as possible.

None of that is true. They are projecting player talent based on their statistical history, often applying considerations like aging curves, and then using those projections to simulate seasons based on how many runs the team will score and concede. That’s it. There’s no ulterior motive and – contrary to popular belief – they aren’t bad. Most do a pretty good job of projecting the final standings, and do so a lot better than most humans. There are always misses, because predicting the future is hard, baseball particularly so.

As for us, I can’t promise we don’t hate your team. Personally, I don’t let it get in the way of my win predictions, but like all people, I’m not bias-free. I can’t help it that I like Mike Trout, Shohei Ohtani & Anthony Rendon.

Why not try it yourself and see if you can beat us, and the projections? Fill out the prediction survey for Banished to the Pen and at the end of the season, I’ll determine the 2021 champion. How hard can it be?

Darius Austin is one of the Founding Four of Bat Flips & Nerds. He is recognised as one of the most respected baseball writers from the UK. Follow him on Twitter @DAustin64 and find his work at Baseball Prospectus and Banished from the Pen

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.