The Good, the Bad, and the Rockfights: What Kind of Team Was Cal in 2023?
Midway through the year the 3-4 Bears were on the edge of The Good. 3 wins and 3 losses later, where did they finish the year?
[Programming update: Nick’s usual Monday column will be delayed due to a power outage resulting from this weekend’s wild winds]
The Cal football season has long passed (and as far as I am concerned, the NFL season might as well be over too), and we are far enough removed from 2023 that we can start to reflect on what kind of team we watched last year. This is the first of two year-end Good, Bad, Rockfight pieces that will (hopefully!) help us better understand the 2023 Cal Bears. This series is usually behind a paywall because it relies on proprietary (read: expensive) data from our friends at Pro Football Focus (PFF). So for those of you unfamiliar with this series, I will give a quick overview. Those already familiar—thanks for subscribing and helping us fund that PFF subscription!
PFF has an army of video junkies who grade every player on every snap for each game, rating everything from tackling to blocking to running to pass coverage. This series uses those grades aggregated at the team level to get a profile of how the Bears performed in each game. Typically we compare each week’s grades and then use machine learning to figure out what “type” of Cal game we watched. So far we have identified four types of Cal games: Good (self-explanatory, most categories were graded well), Bad (another self-explanatory one: most categories were graded poorly), Rockfights (defensive earns excellent grades while the offense tends to be sub-par, especially in the passing game), and Pillowfights (passing game excels while the defense earns poor grades). This time we are not looking at a single game; rather, we’re looking across the entire 2023 season. We did something similar during the bye week and found that, as a whole, the 2023 team belonged in The Good. Since then Cal posted a rather wild finish to the year en route to bowl eligibility—how do things look now over the course of the season?
PFF Grades Comparison
The plot below lays out each of the twelve team-level categories that PFF grades. I have organized the game-level data into boxplots: the box captures data between the 25th and 75th percentiles while the blue horizontal line represents the median. The vertical lines extending beyond each box capture data points above the 75th and below the 25th percentiles, while the blue dots capture extreme outliers. I have superimposed dots over each boxplot showing average grades for each season; they are color-coded by season.
Our primary focus is on the dark gray dots, which represent the 2023 season. Unfortunately, the average grades only fared better-than-usual in three categories: offense, receiving, and running; and only receiving was substantially better than usual. The rest fell below usual, with pass protection, run blocking, run defense, and tackling faring particularly poorly. As poorly as the O-line was graded, it was a big step up from 2022’s nadir. In fact, in all categories except running, the 2023 team was graded 5-10 points higher than the 2022 team. In other words, they have very similar strengths and weaknesses, but the 2023 was simply a better team across the board. That is reflected in the records as the 2022 team finished a dismal 4-8 while 2023 earned two more wins.
Next I took those average grades and fed them into a cluster analysis machine learning algorithm. The algorithm sorts the data into clusters of games that are similar to each other and (ideally) dissimilar from games in other clusters. The plot visualizes those clusters below with color-coding. Here we have our usual set of games: Good (blue), Bad (red), Rockfights (gold), and Pillowfights (green).
The above plot visualizes the four clusters built by our clustering algorithm. Each data point represents a single game (plus our averages for each season) and points closer together earned more similar grades while points farther apart earned more dissimilar grades. There are many ways grades can vary across games, so these games are scattered about in this plotting space (note: unlike plots where the x- and y-axis correspond to some meaningful variable, here the plotting space only serves to illustrate differentiation among data points. You could take this plot and invert the x- or y-axis and the interpretation would still be the same).
In addition to every game from the last seven seasons, this plot also includes season-average grades: those are outlined with rectangles. Interestingly, all the season averages are relatively close to the center of the plot. The reason for this is rather simple: if you take a random sample of games in the plot, they’ll be scattered all over; take their average and it will be somewhere in between them, likely near the middle of the plot. As each year is an average of several games, it’s more likely to be somewhere near the middle of the plot. Despite their migration towards the middle, each year’s average tells us something useful about that year. 2017 falls within The Good while 2018 sits comfortably in The Rockfights. 2018 is farthest from the middle and rather comfortably nestled in that cluster, suggesting that the 2018 team pretty well encapsulates a Rockfight team. By contrast, 2019 sits on the perimeter of The Rockfights—it’s barely clinging onto that category. 2020 is our lone representative in The Bad. Despite the 5-7 record, 2021 finished in The Good, although it also sits on the perimeter; it’s hardly a strong representative of The Good, more like a team that couldn’t really fit anywhere else so it got slapped onto the edge of the category. 2022 is our first Pillowfight season, and it sits somewhat comfortably inside the category. With the worst defensive ratings of the Wilcox Era and some decent offensive ratings, 2022 is a prototypical Pillowfight.
That leads us to 2023, another entry in The Pillowfights. Like many other season averages, it sits near the perimeter of the cluster, so it is not a comfortable fit within there. That is reflected by the variety of game types we saw in 2023: 5 Good (5 wins), 4 Pillowfights (3 losses, 1 win), 2 Bad (2 losses), 2 Rockfights (2 losses). This marks two years in a row where the season-average grade gets classified as a Pillowfight, quite a contrast to the previous years under Wilcox. 2024 provides Cal an opportunity to reset in a new conference and try to restore the defense to the 2017-19 levels. If the Bears can achieve that while maintaining their newfound offensive success, they should be able to achieve a season belonging to The Good. We’ll find out in just under seven months…