Is Bryce Harper a “Bust” This Season? Or is He Just More Consistent?
Here's a post about baseball… but it's really about statistics. Baseball has an endless supply of numbers and time-series data. This post is also about using Process Behavior Charts to evaluate performance over time.
Process Behavior Charts help us stop overreacting to every up and down in a metric, whether that's a baseball player's OPS (on-base percentage plus slugging percentage) or if it's a hospital's patient satisfaction scores.
Sports fans (and talk radio hosts) love overreacting to changes in a metric like OPS. Reacting is easy to do. Higher is better! Lower is worse! Oh, and many business leaders do this too.
But, a more important question is if those changes amount to “noise” (or routine fluctuation) in the metric or if the change is a “signal” that says the player's performance has changed in a significant way or a sustained way… or both.
One such player that people love reacting to is Bryce Harper, who signed an enormous contract with the Philadelphia Phillies this past off season.
I was reminded of him with this headline the other day:
So far in August, Harper has his highest OPS of the season so far.
That much is obvious, even from a table of numbers:
Drawing a Process Behavior Chart with just these six data points looks like this:
A note for those who know about Process Behavior Charts (or other forms of “control charts” or “SPC charts”), yes, you can create a chart with just six (or just four) data points. Tthe Lower and Upper Limits just aren't as valid as they would be with more data points.
There's no signal with the 2019 X Chart… no signal in the companion MR Chart. August is indeed higher, but it's not above the calculated Upper Limit. It's the highest this season, but it's not a signal.
Every metric has a “highest point ever” — that doesn't make it a statistically meaningful signal.Embed from Getty Images
To put the data in
An article from just six days ago was still asking if he's a “bust”:
Instead of taking his word (or the local sports talk loud mouth's word) for it, let's look at the numbers. We can look at Harper's monthly OPS numbers for his entire career. That chart is seen below (using his first three seasons, or 16 data points, as the baseline for calculating the average and limits):
The chart tells us that Harper is an inconsistent player with high highs in some months and low lows.
His average OPS from his first three seasons was 0.812. You can see that his monthly OPS had been fluctuating around the average, but then his 2015 season (when he was the National League MVP) was clearly his best.
In 2015, Harper had two months above the calculated Upper Limit (what I call a “Rule 1” signal) and he had a cluster of three months that were closer to the Upper Limit than they were to the average (what I call a “Rule 3” signal). There were seven consecutive months that were all above average, which was one month away from being a “Rule 2” signal. 2015 was an unusually good season for Harper. The chart shows that much.
We see other signals in later seasons. I'll mark them below:
There are a total of four signals (you could call them “positive outliers”) where his monthly OPS was above the Upper Limit. In those months (and for that entire 2015 season), it would be fair to ask, “What happened? What was different for Bryce Harper?” Those data points are statistically unlikely to be random variation.
There are two grey boxes that show the Rule 3 signals. Harper had two positive outliers in the 2017 season but then September 2017 was his worst month ever and that OPS data point is just below the Lower Limit. What was different that month?? That's a month worth asking about. Was he injured?
Every data set has a “worst month ever” — and it could be a signal. The three rules tell us if we have a signal:
Looking at the MR Chart, there are two signals — the first caused by a huge INCREASE in OPS and the second caused by a huge DECREASE in OPS from month to month. Those are worth asking about — when we wouldn't ask why there were small ups and downs in the 2014 season.
Now, the last eight months for Harper, including the first five months of this year, appear to be above his baseline average. It's ALMOST a “Rule 2”
We could treat that as a signal… and I'd answer the question of “Is Bryce Harper a Bust?” like this:
- If you thought you were going to get 2015 Bryce Harper every year, you'd be mistaken… that season was an outlier for him.
- If anything, Harper is performing at an “above-average” rate this season and Phillies fans (and those who talk about the game) shouldn't panic. I would have drawn that conclusion before his high August month, even.
- His performance might even be shifting upward a bit… and getting more consistent than his earlier seasons.
It's possible that we could shift the limits so that his future performance is predicted to be within this slightly higher, more consistent (less variable) range:
Even though the “bust” word was being thrown around, the first months of 2019 were above average for Harper.
As this headline from late April said:
And he is. And he was, even in April.
“Harper has gone into deep slumps any number of times over the course of his previous seven big-league seasons. He came out of them. He will come out of this one. When he does, and that is likely to happen soon, he can carry this team for two weeks.”
There's going to be variation in any metric. There's variation in any system. You wouldn't expect Bryce Harper (or any ballplayer) to have the same stats every month.
What did Harper's own words from that one article say?
“I'm not going to tell you I'm going to win MVP every single year. … There's going to be down years, there's going to be big years, there's going to be years that are just OK.”
Harper is explaining variation to us. He has such great talent, his high seasons are going to be phenomenally good. Does he have more of them ahead? Only time will tell.
If we used his ENTIRE career to calculate the average and limits, the chart looks like this:
We see the 2015 outlier data point… the point above the Upper Limit and the near-signal of seven consecutive above his career average. We now see nine consecutive below-average months early in his career. September 2017 was a negative outlier still.
But the story about this season and recent months is similar to the other PBC that used his first three seasons as the baseline:
- Four consecutive below-average months is not a signal or a trend (he wasn't a bust)
- This year, he's fluctuating around his career average and will likely continue to do so (yeah, calm down)
- The question remains — will he ever have positive outlier months or MVP seasons again? That's the $330 million gamble made by the Phillies.
Comparisons to Mike Trout
If we look at the career of Mike Trout, a very comparable baseball superstar, we see that Trout is more consistent… but, again, there's always variation:
Trout has a much higher baseline average OPS at 0.984 (calculated from the first 24 months of his career).
The only signal there was the first month of his career… beyond that, he's predictable… within a range. Ten of his last eleven months are all above average, so that could mean that Trout's performance is shifting upward a bit, as is Trout's. It's not technically a Rule 2 signal, but it also doesn't look like random fluctuation. It's scary, but Mike Trout might be getting better.
Trout has higher performance… and more variation… but it's more consistent variation compared to Harper.Embed from Getty Images
Here's my spreadsheet with Harper and Trout analysis (via Dropbox).
Comparing to Albert Pujols
It's perhaps more difficult to predict future performance in sports because the system is changing in various ways… players get older, teams change their pitching and defensive strategies against a player (such as employing defensive shifts more often).Embed from Getty Images
A Process Behavior Chart showing the yearly OPS for the great Albert Pujols shows a marked shift downward starting in 2011 or so.
From the first ten seasons of his career, he had an average OPS of 1.050. That stat, sustained for an entire career, would have put him in the top ten of all time (along with Mike Trout). But he's now dropped to 38th of all time — still in the company of Hall of Famers (which he will be).
The PBC would have predicted that his OPS would continue fluctuating between the calculated Upper and Lower Limits. But, some blame the defensive shift for changing the system, making his performance change from the previously predictable pattern:
That change to the system (the infield shift) led to an apparent SHIFT downward in his OPS performance… that seems to be a clear (or believable) cause-and-effect relationship.
We'd no longer predict that his OPS would fluctuate around 1.000… it's been fluctuating around 0.750 in recent seasons. Again, he's probably still a Hall of Famer… but the decline is noticeable in the chart… not a steady decline, but a step function downward. The Albert Pujols of the last nine seasons is not getting the same results as the Albert Pujols of his first ten seasons.
His performance would likely remain in that lower range, unless MLB bans the defensive infield shift (as they're piloting in the Atlantic League this year). Pujols would have wished they made that change to the system a decade ago. Then again, the article quotes Pujols talking about how he refuses to change his approach to hitting… when the system around you changes, sometimes you have to change too…
Here is the Pujols data and my spreadsheet if you want to play around with it.
What are your thoughts about any of these players and how this analysis applies to some of your workplace metrics?